1 00:00:07,799 --> 00:00:11,160 Speaker 1: Welcome back to Adventures in DevOps. I'm your host, Warren 2 00:00:11,320 --> 00:00:14,080 and because today Will is away this week, I have 3 00:00:14,119 --> 00:00:17,160 an opportunity to sneak in a sponsorship. Today's episode is 4 00:00:17,160 --> 00:00:20,359 sponsored by Attribute. I actually met the team and honestly, 5 00:00:20,440 --> 00:00:23,519 what they're doing in the Finnoffs space is absolutely genius 6 00:00:23,879 --> 00:00:26,399 that I believe actually everyone can benefit from. They call 7 00:00:26,440 --> 00:00:30,239 it finops without tagging. It's the first run time technology 8 00:00:30,320 --> 00:00:33,920 that analyzes infrastructure instead of relying on billing reports, exports 9 00:00:34,039 --> 00:00:38,119 and tagging. It's for architecture ops and platform teams that 10 00:00:38,159 --> 00:00:42,039 need visibility into product customer attribution or insight into cost 11 00:00:42,079 --> 00:00:45,960 anomalies without wasting hours of guessing how to allocate spend 12 00:00:45,960 --> 00:00:49,880 too shared services. So there's no spreadsheets, no extra logging. 13 00:00:50,159 --> 00:00:52,719 Attribute solves it all with just one line of code. 14 00:00:53,119 --> 00:00:57,640 They capture costs based on actual application usage generated from 15 00:00:57,719 --> 00:01:02,000 anywhere Kubernetes, databases, Story and over thirty five multi cloud 16 00:01:02,079 --> 00:01:04,599 services in there UI. They break it down by micro 17 00:01:04,680 --> 00:01:07,680 service and even at tribute cost at the database query level, 18 00:01:08,000 --> 00:01:08,799 all tied back. 19 00:01:08,640 --> 00:01:09,200 Speaker 2: To the business. 20 00:01:09,239 --> 00:01:13,040 Speaker 1: I really find that pretty interesting. Recently they were recognized 21 00:01:13,040 --> 00:01:15,400 by six Gardner hype cycles. I honestly have no idea 22 00:01:15,439 --> 00:01:17,760 what that is though, and are working with impressive companies 23 00:01:17,760 --> 00:01:20,920 like Akamian Monday dot com, so you'll want to check 24 00:01:20,959 --> 00:01:22,959 them out. And I'll drop a link in the description 25 00:01:23,040 --> 00:01:26,120 for the episode and that's attribute at ATTRB dot io. 26 00:01:26,239 --> 00:01:30,519 And now back to the show and today, I have 27 00:01:30,599 --> 00:01:34,200 to say I'm actually pretty intrigued by the guests that 28 00:01:34,239 --> 00:01:36,680 we brought on because this is an area of technology 29 00:01:36,680 --> 00:01:38,840 that I have zero experience with. We're going to be 30 00:01:38,840 --> 00:01:41,680 talking all about vector databases, and I feel like we 31 00:01:41,719 --> 00:01:44,560 brought in one of the experts from the industry, from 32 00:01:44,560 --> 00:01:46,879 a company that has been doing vector of databases for 33 00:01:46,959 --> 00:01:49,400 quite some time now, I want to say since the beginning, 34 00:01:49,400 --> 00:01:51,640 but I think she's going to correct me. So welcome 35 00:01:51,680 --> 00:01:55,239 to the show. Staff Developer Relations Jenna Patterson. 36 00:01:55,439 --> 00:01:57,599 Speaker 3: Hello, thank you for having me on today. 37 00:01:57,879 --> 00:02:00,400 Speaker 1: Yeah, I know, I'm really interested because actually, and part 38 00:02:00,400 --> 00:02:03,319 of preparing for this episode, I went around and asked 39 00:02:03,680 --> 00:02:06,120 a lot of my colleagues at different companies if they 40 00:02:06,200 --> 00:02:09,840 had any questions that I should ask someone who's I mean, 41 00:02:09,879 --> 00:02:12,199 you sort of just got into the work at pine 42 00:02:12,199 --> 00:02:14,400 Cone there just under the year. If I I'm right, 43 00:02:15,280 --> 00:02:17,400 what I should ask them and they're like, I don't 44 00:02:17,400 --> 00:02:18,000 know what that is. 45 00:02:18,280 --> 00:02:21,319 Speaker 4: Yeah, So high level it allows you to do to 46 00:02:21,439 --> 00:02:25,960 compare vector emvetting numerical representation of a piece of data, 47 00:02:26,120 --> 00:02:28,759 and so it's and the vector database allows you to 48 00:02:28,759 --> 00:02:34,319 find similar matches. So if you think about searching on 49 00:02:35,039 --> 00:02:38,360 for instance, an e commerce store retail store online, and 50 00:02:38,400 --> 00:02:40,479 you want to find all the shirts that are read, 51 00:02:40,680 --> 00:02:43,280 they might not all have the word shirt in the description. 52 00:02:43,439 --> 00:02:46,719 So we want to find everything that is related or 53 00:02:46,800 --> 00:02:49,639 closely related, the most closely related. And so it does 54 00:02:49,680 --> 00:02:55,000 that based on a distance metric to find everything that 55 00:02:55,280 --> 00:02:58,680 has a meaning similar to shirt or in this case, 56 00:02:58,879 --> 00:03:01,080 a red shirt could be a blow so it could 57 00:03:01,080 --> 00:03:03,719 be a top, it could be a short sleeve shirt. 58 00:03:04,439 --> 00:03:07,840 And then it comes back as scored results, so that 59 00:03:07,960 --> 00:03:10,439 like anything that is close is going to have a 60 00:03:10,479 --> 00:03:13,280 better score than the things that are further away from 61 00:03:13,280 --> 00:03:14,599 that particular query. 62 00:03:14,879 --> 00:03:15,240 Speaker 2: I see. 63 00:03:15,280 --> 00:03:18,599 Speaker 1: So you take the original requests red shirt, and you're 64 00:03:18,599 --> 00:03:21,319 converting it to some just a set of numbers and 65 00:03:21,360 --> 00:03:23,800 then using that to look through the database and get 66 00:03:23,840 --> 00:03:28,280 back equivalent results. So there's some upfront converting being done. 67 00:03:28,319 --> 00:03:31,280 I assume when you're sticking data into the database in 68 00:03:31,400 --> 00:03:35,560 order to store the numerical representations, not just the raw 69 00:03:35,840 --> 00:03:37,199 say text properties. 70 00:03:37,439 --> 00:03:38,280 Speaker 3: Yes, exactly. 71 00:03:38,400 --> 00:03:40,960 Speaker 4: So at the beginning, you're going to ingest all of 72 00:03:41,000 --> 00:03:43,280 your data, so you chunk it up, and then you 73 00:03:43,360 --> 00:03:44,840 upsort it into a vector. 74 00:03:44,599 --> 00:03:46,439 Speaker 3: Database ahead of time. 75 00:03:46,840 --> 00:03:50,560 Speaker 4: As your users are querying, we take that query, we 76 00:03:50,680 --> 00:03:52,840 embed that query and use that as a way to 77 00:03:53,000 --> 00:03:57,120 compare against those existing embeddings that are in the database. 78 00:03:57,319 --> 00:03:58,199 Speaker 3: There's also kind of. 79 00:03:58,159 --> 00:04:02,840 Speaker 4: Another piece about that where if that data changes or 80 00:04:02,879 --> 00:04:04,919 you get new data, you can also reupsert that. 81 00:04:05,159 --> 00:04:05,560 Speaker 2: I see. 82 00:04:05,599 --> 00:04:08,080 Speaker 1: So I mean this I assume is true for all 83 00:04:08,400 --> 00:04:11,280 databases that claim that they're vector databases. 84 00:04:11,439 --> 00:04:13,439 Speaker 2: How do you do the computation. 85 00:04:13,120 --> 00:04:15,960 Speaker 1: Of figuring out what the numerical value should be for 86 00:04:16,519 --> 00:04:17,040 red shirt? 87 00:04:17,240 --> 00:04:19,519 Speaker 4: We do that through what is called an embedding model. 88 00:04:19,600 --> 00:04:22,519 But this embedding model is traded on specific data that 89 00:04:22,680 --> 00:04:26,639 is for embedding, and we pass in the text value. 90 00:04:26,680 --> 00:04:28,759 So that chunk of data or a piece of text 91 00:04:28,920 --> 00:04:32,000 in this case, it's text, it spits out some numbers 92 00:04:32,040 --> 00:04:34,279 and it happens to be in vector form. So if 93 00:04:34,279 --> 00:04:36,240 you remember back to like I don't know, fifth or 94 00:04:36,279 --> 00:04:41,040 seventh grade geometry, we worked with vectors essentially a list 95 00:04:41,079 --> 00:04:44,480 of numbers. But these are very very long vectors, so 96 00:04:44,600 --> 00:04:48,959 very high dimensional data onenty twenty four dimensions in this vector, 97 00:04:49,519 --> 00:04:54,360 and these represent different pieces of meaning about that piece 98 00:04:54,360 --> 00:04:57,560 of data. So it could be about the color. It 99 00:04:57,600 --> 00:05:00,800 could be in this case, like we're outside of the query, 100 00:05:01,000 --> 00:05:04,519 like the data we might embed, like the product description, 101 00:05:04,639 --> 00:05:07,759 we might embed product title, or all of that together. 102 00:05:07,839 --> 00:05:09,879 That could be part of our chunking strategies to put 103 00:05:09,920 --> 00:05:12,120 it all together and embed that whole thing as one piece. 104 00:05:12,560 --> 00:05:14,439 Speaker 1: I just have like a lot of questions now, like 105 00:05:15,000 --> 00:05:17,759 first and foremost, is it your fault that when I 106 00:05:17,800 --> 00:05:19,800 search for red shirt on websites? 107 00:05:19,839 --> 00:05:21,560 Speaker 2: I find things that aren't red shirts? 108 00:05:21,639 --> 00:05:24,279 Speaker 3: Now, I hope it's not our fault. 109 00:05:25,000 --> 00:05:29,439 Speaker 1: If I understand you correctly, that actually the model I 110 00:05:29,480 --> 00:05:32,399 assume you're using like something similar to an LM to 111 00:05:32,480 --> 00:05:37,040 convert from the original text into what the embedded embedding 112 00:05:37,079 --> 00:05:39,439 value should be that you're storing in the database. You're 113 00:05:39,439 --> 00:05:41,439 doing that because pine Cone. 114 00:05:41,160 --> 00:05:42,160 Speaker 2: Has this capability. 115 00:05:42,160 --> 00:05:44,279 Speaker 1: But if you're using let's say one of your I 116 00:05:44,319 --> 00:05:46,319 don't see one of your competitors, but an open source 117 00:05:46,519 --> 00:05:49,680 vector database or I think postcress supports vectors now that 118 00:05:49,839 --> 00:05:53,360 responsibility is on the implementer of or the team that 119 00:05:53,480 --> 00:05:55,360 is actually implementing searching their application. 120 00:05:55,480 --> 00:05:59,040 Speaker 4: Right. Yeah, So there's a couple of different approaches to 121 00:05:59,120 --> 00:06:01,879 it at Pine, and we have two different approaches. We 122 00:06:01,959 --> 00:06:05,199 support two different ways. So you might actually have your 123 00:06:05,199 --> 00:06:08,639 own embeddings already. You might want to manage that part 124 00:06:08,639 --> 00:06:12,199 of the process yourself, and so you might use something 125 00:06:12,279 --> 00:06:15,839 like an open AI model to do embedding or an 126 00:06:15,839 --> 00:06:18,959 Amazon model to do embedding. We also host our own models. 127 00:06:19,079 --> 00:06:21,680 It's an Nvidia model for embedding. We have a number 128 00:06:21,720 --> 00:06:24,360 of those based on your use case. And right now 129 00:06:24,360 --> 00:06:27,439 we're talking about text like product descriptions, product titles. It 130 00:06:27,439 --> 00:06:30,959 could be audio, it could be images, any sort of 131 00:06:31,040 --> 00:06:33,720 data that you want to actually do like a meaningful 132 00:06:33,759 --> 00:06:37,800 search over and find meaning as opposed to specific keywords. 133 00:06:38,079 --> 00:06:41,399 Speaker 1: How is the comparison being done like in the database side, 134 00:06:41,439 --> 00:06:42,920 Like I get the part where you run through an 135 00:06:42,920 --> 00:06:45,759 embedding model and you get back out an array of 136 00:06:46,000 --> 00:06:50,720 onenty twenty four integers or floating point numbers that somehow 137 00:06:50,720 --> 00:06:52,199 gets stored in a database. 138 00:06:52,680 --> 00:06:55,519 Speaker 2: Is like is it being stored as like a row value? 139 00:06:55,639 --> 00:06:58,879 Speaker 1: Is there like a special format that the database saves 140 00:06:59,000 --> 00:07:02,279 data in is this like an incredibly complicated question to answer? 141 00:07:02,639 --> 00:07:05,319 Speaker 4: This is for me, it's an incredibly complicated question to answer, 142 00:07:05,360 --> 00:07:07,240 but I can answer part of it. So the comparison 143 00:07:07,279 --> 00:07:09,959 that's being done is that we are taking those two 144 00:07:10,000 --> 00:07:13,000 vectors and seeing how far away they are from each other. 145 00:07:13,040 --> 00:07:16,319 So if you think of a vector like this representing 146 00:07:16,480 --> 00:07:19,000 you know, the red shirt, and a vector going like 147 00:07:19,120 --> 00:07:23,879 this representing like pants, but maybe you have something a 148 00:07:23,879 --> 00:07:26,319 little bit closer. It's really hard to do this as 149 00:07:26,360 --> 00:07:29,439 a podcast and just with my hands, but visuals are 150 00:07:29,519 --> 00:07:30,279 much better. 151 00:07:30,360 --> 00:07:31,560 Speaker 2: For those of you who are just listening. 152 00:07:31,720 --> 00:07:35,040 Speaker 1: Jenna is attempting to draw vectors with her arms, and 153 00:07:35,120 --> 00:07:37,639 if that somehow makes sense, but I totally get. Like, 154 00:07:37,680 --> 00:07:40,560 you have a triangle of vectors and you're calculating the 155 00:07:41,000 --> 00:07:43,920 difference between those vectors from each other, and that's how 156 00:07:43,959 --> 00:07:45,879 you can I'm your calculating distance, and I assume you're 157 00:07:45,879 --> 00:07:48,560 optimizing for the smallest distances possible. One of the questions 158 00:07:48,600 --> 00:07:50,519 I have is like, you have a lot of these 159 00:07:50,560 --> 00:07:55,480 embedding vectors in your database, and aside from how they're 160 00:07:55,480 --> 00:07:58,240 actually being stored, you still have to fetch some set 161 00:07:58,319 --> 00:08:01,199 of that data optimally rather than fetching like all the 162 00:08:01,279 --> 00:08:04,199 data I assume, rather than catching the entire database in 163 00:08:04,319 --> 00:08:07,319 order to compare each vector one by one. Any thoughts 164 00:08:07,319 --> 00:08:10,800 of like how you are able to pare down the 165 00:08:10,839 --> 00:08:13,399 total amounts that you're only fetching relevant data in or 166 00:08:13,439 --> 00:08:14,279 to do that comparison. 167 00:08:14,439 --> 00:08:18,519 Speaker 4: This is definitely beyond my knowledge. I will say, like, 168 00:08:18,800 --> 00:08:22,600 there are strategies for how it's being stored and how 169 00:08:22,639 --> 00:08:23,720 close the data is. 170 00:08:23,959 --> 00:08:25,879 Speaker 1: It's like the thing where I definitely like try to 171 00:08:25,920 --> 00:08:28,759 pose this to people who come up with new database formats, 172 00:08:28,839 --> 00:08:29,480 because there's. 173 00:08:29,279 --> 00:08:30,439 Speaker 2: Always some in this space. 174 00:08:30,600 --> 00:08:33,440 Speaker 1: And when I asked them about it, they were like, oh, yeah, 175 00:08:33,440 --> 00:08:35,279 we build a new database format. And I'm like, well yeah, 176 00:08:35,320 --> 00:08:37,919 but really, what you did was just use your underlying 177 00:08:38,000 --> 00:08:40,879 database engine and you just put a service on top 178 00:08:40,919 --> 00:08:42,840 of it, and you're calling it a database, like it's 179 00:08:42,879 --> 00:08:43,919 really not, at the end of the day, is just 180 00:08:43,960 --> 00:08:46,559 a relational database. What I really understand here is that 181 00:08:46,600 --> 00:08:49,159 it's fundamentally different how you're storing the data. It's not 182 00:08:49,240 --> 00:08:53,720 in arbitruarrily row based information or you know, binary blobs 183 00:08:53,720 --> 00:08:56,000 that are being stored which can be fetched. Fundamentally, the 184 00:08:56,159 --> 00:08:58,799 vector database is storing data not only in a different way, 185 00:08:58,840 --> 00:09:01,799 but has to be optimized in order to find locality 186 00:09:02,200 --> 00:09:05,519 of these vectors and the ones that are close in distance. 187 00:09:05,639 --> 00:09:06,960 Speaker 2: I'm not a database expert. 188 00:09:06,799 --> 00:09:09,879 Speaker 1: So honestly, you know, what you shared so far is 189 00:09:10,399 --> 00:09:11,840 still find find by me. 190 00:09:11,879 --> 00:09:13,480 Speaker 2: But I'm sure someone will call me out on it. 191 00:09:13,559 --> 00:09:15,799 Speaker 4: I do think it's interesting in that, like it's fun 192 00:09:15,840 --> 00:09:19,399 to understand, like the under underlying technology, and I'm like, 193 00:09:19,519 --> 00:09:21,480 as someone who has been here for six months, I'm 194 00:09:21,519 --> 00:09:25,879 still learning that it's very complicated. Even our customers are 195 00:09:25,919 --> 00:09:28,480 like they're learning it along with us as well, like 196 00:09:28,559 --> 00:09:30,559 and like kind of what the strategies are so that 197 00:09:30,559 --> 00:09:33,000 they can implement it in the in the best possible way. 198 00:09:33,120 --> 00:09:35,039 Speaker 2: Right, I think this is the right word. Uh. 199 00:09:35,399 --> 00:09:37,879 Speaker 1: If you have an application where historically you would have 200 00:09:37,879 --> 00:09:40,879 done something like free tech searching in elastic search, using 201 00:09:40,919 --> 00:09:45,080 an embedding model to calculate the numerical values to handle 202 00:09:45,120 --> 00:09:49,720 a semantic search is just the progression in the industry. 203 00:09:49,759 --> 00:09:51,120 Speaker 2: You no longer need free tech search. 204 00:09:51,240 --> 00:09:52,679 Speaker 1: This is the I don't want to say be all 205 00:09:52,759 --> 00:09:56,120 end all of the future of e commerce websites, but 206 00:09:56,440 --> 00:09:59,120 it does seem like there's just such an improvement in 207 00:09:59,159 --> 00:10:01,559 the strategy here from what was done before. 208 00:10:01,639 --> 00:10:03,840 Speaker 4: It is not necessarily that you're not going to use 209 00:10:04,279 --> 00:10:06,879 like a keyword search. You might actually pair them together, 210 00:10:06,919 --> 00:10:09,120 and I'll talk about that in a second. And then 211 00:10:09,399 --> 00:10:12,360 like e commerce is just a it's a simple example. 212 00:10:12,399 --> 00:10:15,159 It's the use case I typically start out with. But 213 00:10:15,240 --> 00:10:18,279 there are other other reasons why you would use this 214 00:10:18,440 --> 00:10:21,399 type of search, semantic search, for instance, in your AI 215 00:10:21,480 --> 00:10:25,159 applications where you are getting you're maybe you're chatting back 216 00:10:25,200 --> 00:10:29,159 and forth with with a model, but the model doesn't 217 00:10:29,240 --> 00:10:31,519 it has its own limitations as we as we know now, 218 00:10:31,600 --> 00:10:34,600 Semantic search is about the meaning behind your query and 219 00:10:34,639 --> 00:10:38,360 your your intent behind what you're trying to find based 220 00:10:38,360 --> 00:10:42,120 on you know what what data is in what context 221 00:10:42,240 --> 00:10:46,679 the data is in. But sometimes you have you know, keywords, 222 00:10:47,120 --> 00:10:52,200 or you have domain specific language or acronyms or stock 223 00:10:52,240 --> 00:10:54,720 tickers is another common one that that we use as 224 00:10:54,720 --> 00:10:58,120 an example. But within your company, you have product names, 225 00:10:58,679 --> 00:11:04,320 you have your own company specific language, you have technical 226 00:11:04,879 --> 00:11:07,639 technical terms that might not be in the public domain 227 00:11:07,679 --> 00:11:10,120 and might not be trained into those models. We might 228 00:11:10,159 --> 00:11:12,480 pair a semantic search with what is called a lexical 229 00:11:12,519 --> 00:11:14,960 search or a keyword search in order to make those 230 00:11:15,000 --> 00:11:17,639 results even more accurate, even more correct. 231 00:11:17,720 --> 00:11:19,679 Speaker 1: I think you really stumble onto a good thing here 232 00:11:20,000 --> 00:11:22,720 that's worth talking about, because I think where you're going 233 00:11:22,879 --> 00:11:24,559 is for those of you who are unfamiliar or may 234 00:11:24,600 --> 00:11:26,720 have heard of RAG before, I think it really jumped 235 00:11:26,799 --> 00:11:30,799 jumping onto the retrieval augmented generation where fundamentally, if you 236 00:11:30,799 --> 00:11:33,720 think about it, the models that you're using that are 237 00:11:33,759 --> 00:11:38,039 proprietary by third party companies don't understand how you talk 238 00:11:38,080 --> 00:11:41,000 about your business, and so how do you do the 239 00:11:41,039 --> 00:11:45,240 mapping from one of these models to what's internally? And 240 00:11:45,600 --> 00:11:48,840 what you're saying is if you've uploaded all the data 241 00:11:49,000 --> 00:11:52,440 or you have hooks into your knowledge bases, and you 242 00:11:53,519 --> 00:11:56,720 throw that through an embedding model into pine Cone or 243 00:11:56,720 --> 00:12:01,320 another vector database, then using the map of MCP or 244 00:12:01,440 --> 00:12:05,240 some other magic that no one knows about, somehow the 245 00:12:05,279 --> 00:12:08,840 model gets access to this data and understands how to 246 00:12:09,440 --> 00:12:14,000 use the embedding values from their own model to map 247 00:12:14,080 --> 00:12:17,840 to what's internal because there is a semantic likeness between 248 00:12:17,879 --> 00:12:20,600 those things, even though fundamentally the actual words and even 249 00:12:21,039 --> 00:12:24,000 outside and say a dictionary, those things are fundamentally different. 250 00:12:24,200 --> 00:12:25,120 Speaker 3: Yeah, exactly. 251 00:12:25,200 --> 00:12:27,240 Speaker 4: I think one way I like to look at it 252 00:12:27,320 --> 00:12:29,919 is like we you and I, Like we have a 253 00:12:29,960 --> 00:12:30,679 spoken language. 254 00:12:30,720 --> 00:12:31,519 Speaker 3: We speak English. 255 00:12:31,639 --> 00:12:34,240 Speaker 4: It's natural to us, and so like, if we can 256 00:12:34,279 --> 00:12:37,399 interact with our data and gain insights and find information 257 00:12:38,200 --> 00:12:41,080 using the language we know the best, that's going to 258 00:12:41,080 --> 00:12:44,240 be even better for our output. Lllms have limitations and 259 00:12:44,320 --> 00:12:46,879 so and part of that is it doesn't know all 260 00:12:46,919 --> 00:12:49,480 about our data. And so if we bring in the knowledge, 261 00:12:49,519 --> 00:12:55,000 bring in the factual data, the authoritative data into that process, 262 00:12:55,080 --> 00:12:57,360 then are our output can be even better. 263 00:12:57,480 --> 00:12:59,200 Speaker 1: You may be the first person I've talked to that 264 00:12:59,320 --> 00:13:03,720 suggested that English was like a good strategy for communication. 265 00:13:03,879 --> 00:13:06,759 Speaker 3: Here, I don't know if I'm suggesting it's a good strategy. 266 00:13:06,840 --> 00:13:08,720 Speaker 2: I think what can be used. 267 00:13:09,759 --> 00:13:12,679 Speaker 1: I mean, there's just so much that's not shared as 268 00:13:12,759 --> 00:13:16,279 far as the as far as the context goes. Yeah, 269 00:13:16,320 --> 00:13:18,919 that is just fundamentally lost. And I can see how 270 00:13:19,120 --> 00:13:21,960 a trouble and troubling it is to actually communicate in 271 00:13:21,960 --> 00:13:24,519 that way, because we each have our own internal view 272 00:13:24,519 --> 00:13:28,000 of the world. And I know, as a technologist, a 273 00:13:28,080 --> 00:13:31,879 long time technologist, a non tribal, amount of my conversations 274 00:13:31,919 --> 00:13:34,879 have pivoted from talking about whatever the topic is to 275 00:13:35,000 --> 00:13:37,240 like a meta level, like what are we actually talking 276 00:13:37,240 --> 00:13:40,200 about here? You know, let's define some of these words. 277 00:13:40,480 --> 00:13:42,840 It was a recent conversation I was having about the 278 00:13:42,879 --> 00:13:47,039 effectiveness of feature flagged and what came up started with like, well, 279 00:13:47,039 --> 00:13:50,200 how do you use them? Evolved into is it good 280 00:13:50,240 --> 00:13:52,559 to use them? And even if I say that, it's 281 00:13:52,559 --> 00:13:55,159 like what does good mean? And one of the terms 282 00:13:55,200 --> 00:13:57,000 that came up was oh, yeah, you know if you 283 00:13:57,080 --> 00:13:58,960 have everything that's fully tested, And I'm like, what does 284 00:13:59,000 --> 00:13:59,919 fully tested mean? 285 00:14:00,320 --> 00:14:01,799 Speaker 2: Like how do you actually define that? 286 00:14:01,840 --> 00:14:03,600 Speaker 1: In that regard, I feel like it's a very well 287 00:14:03,799 --> 00:14:06,320 I know when I see it, but defining it is 288 00:14:06,360 --> 00:14:08,200 like hugely problematic. 289 00:14:07,879 --> 00:14:08,720 Speaker 3: Super hard. Yeah. 290 00:14:08,840 --> 00:14:11,600 Speaker 4: I remember back in one of my college courses a 291 00:14:11,639 --> 00:14:14,840 long long time ago where we talked exactly about this, 292 00:14:14,960 --> 00:14:18,000 and like how you define a specification that is that 293 00:14:18,080 --> 00:14:23,559 everyone understands and could potentially be executed? Right, So it's 294 00:14:23,559 --> 00:14:25,960 not just that you and I can talk about it, 295 00:14:26,000 --> 00:14:29,039 but that our computer understands it. So I think you 296 00:14:29,080 --> 00:14:31,799 bring up a really good point. I think you're right 297 00:14:31,879 --> 00:14:35,080 in that, like it depends on who we are and 298 00:14:35,120 --> 00:14:37,039 what our experiences are and where we come from and 299 00:14:37,360 --> 00:14:39,559 what language we speak and there's definitely going to be 300 00:14:40,120 --> 00:14:42,360 some of that that happens, and I think there are 301 00:14:42,399 --> 00:14:44,840 ways around that, and I will say right now, I 302 00:14:44,840 --> 00:14:46,399 don't know all of those ways around that. 303 00:14:46,600 --> 00:14:49,240 Speaker 1: Something that I think frequently comes up on the show 304 00:14:49,480 --> 00:14:53,399 is the mentioning that each new release of a model, 305 00:14:53,440 --> 00:14:58,200 public or proprietary is like having another child where they're 306 00:14:58,360 --> 00:15:01,720 like so fundamentally different, not like an upgrade is fun. 307 00:15:02,480 --> 00:15:04,000 How and so I think you brought this up a 308 00:15:04,039 --> 00:15:07,720 little bit. Is the idea where if you're upgrading your 309 00:15:08,000 --> 00:15:10,919 I say upgrading side grading your model for one reason 310 00:15:11,000 --> 00:15:15,240 or another, that revalidating the embeddings that you got previously 311 00:15:15,840 --> 00:15:19,080 match was coming out of the new model to ensure 312 00:15:19,080 --> 00:15:22,279 that you aren't just going to start getting nonsensical outputs. 313 00:15:22,279 --> 00:15:23,799 I mean, in some way you're upgrading the model because 314 00:15:23,799 --> 00:15:26,200 you think the new embeddings will be better, but that 315 00:15:26,279 --> 00:15:29,600 obviously has an opportunity for unexpected results. Does Pinecone have 316 00:15:29,639 --> 00:15:31,240 a strategy for dealing with that? 317 00:15:31,279 --> 00:15:31,720 Speaker 2: I mean, I. 318 00:15:31,639 --> 00:15:33,720 Speaker 1: Assume if you have an embedding model, given what the 319 00:15:33,720 --> 00:15:36,240 company is doing, you're building the model yourself. 320 00:15:36,320 --> 00:15:38,639 Speaker 4: The Nvidia model that one that I mentioned, that's the 321 00:15:38,679 --> 00:15:39,720 hosted model that we have. 322 00:15:40,000 --> 00:15:41,279 Speaker 3: We host it internally. 323 00:15:41,960 --> 00:15:44,919 Speaker 4: We also have our research team and they have created 324 00:15:45,000 --> 00:15:47,120 models for us, so we have the one that I 325 00:15:47,120 --> 00:15:50,360 am most familiar with is Pinecone. Sparse is used for 326 00:15:50,759 --> 00:15:55,039 a lexical search keyword search with sparse vectors as opposed 327 00:15:55,039 --> 00:15:57,440 to dense vectors, which are used for semantic search. 328 00:15:58,240 --> 00:16:00,320 Speaker 2: What's the difference between In a. 329 00:16:00,320 --> 00:16:03,240 Speaker 4: Dense vector, you have a vector of numbers. They represent 330 00:16:03,360 --> 00:16:07,399 different parts of meaning about that particular piece of data 331 00:16:07,440 --> 00:16:11,519 that was embedded. But with a sparsevector, you have more 332 00:16:11,600 --> 00:16:14,399 zeros than you have actual numbers, and it is either 333 00:16:14,440 --> 00:16:18,360 a zero or a one that essentially represents the frequency 334 00:16:18,399 --> 00:16:19,559 of our particular word. 335 00:16:19,799 --> 00:16:22,600 Speaker 1: And I think, I think realistically, if you no one wants, 336 00:16:22,639 --> 00:16:25,080 no one wants to listen to this the you know, 337 00:16:25,120 --> 00:16:27,519 you think back to when you had you went further 338 00:16:27,759 --> 00:16:32,600 and you're like, uh, normalizing arrays where you're you create 339 00:16:32,639 --> 00:16:35,960 some sort of orthonormal basis for actually moving the data 340 00:16:36,000 --> 00:16:38,240 out so that you end up with these arrays where 341 00:16:38,480 --> 00:16:41,480 you have just an amplitude at one position in the vector, 342 00:16:41,639 --> 00:16:45,480 which makes it much easier to identify things in close proximity. 343 00:16:45,759 --> 00:16:48,600 So you can imagine that it's you're not just letting 344 00:16:48,639 --> 00:16:52,080 the embedding model come up with arbitrary numbers to represent 345 00:16:52,360 --> 00:16:56,320 your semantic text, but you're then applying some clever mathematics 346 00:16:56,320 --> 00:16:58,759 on top of it to organize the vectors in a 347 00:16:58,759 --> 00:17:02,000 way so that when you actually go and do a search, 348 00:17:02,120 --> 00:17:04,759 you're not having to pull out every single piece of 349 00:17:04,799 --> 00:17:08,960 data from the database. So being inpecific about how you 350 00:17:08,960 --> 00:17:12,240 can utilize the mathematics who optimize your vector database goes 351 00:17:12,279 --> 00:17:14,200 into some of the creation of these. So a lot 352 00:17:14,240 --> 00:17:16,559 of engineers got into i'd say software engineering or became 353 00:17:16,599 --> 00:17:19,480 engineers in the first place because they told themselves and 354 00:17:19,480 --> 00:17:21,359 I'm going to say a lie that they wanted to 355 00:17:21,400 --> 00:17:24,640 work on hard problems. And it does actually sound like 356 00:17:24,920 --> 00:17:28,599 that building like compared to building up other databases, I 357 00:17:28,680 --> 00:17:31,720 do feel like understanding the mathematics behind a vector database 358 00:17:31,799 --> 00:17:32,400 is non trivial. 359 00:17:32,960 --> 00:17:33,799 Speaker 3: Yeah, I agree. 360 00:17:33,920 --> 00:17:36,240 Speaker 4: I am that person who, like I want to work 361 00:17:36,279 --> 00:17:38,559 on hard problems. I like knowing how things work, and 362 00:17:38,599 --> 00:17:42,559 so again I'm still learning a lot of this obviously. 363 00:17:42,759 --> 00:17:45,680 I think for me that is one of the fascinating 364 00:17:45,720 --> 00:17:47,480 pieces about it is that it's math. 365 00:17:47,599 --> 00:17:48,640 Speaker 3: It's not easy math. 366 00:17:48,720 --> 00:17:50,720 Speaker 4: I mean, like we go to school for it, and 367 00:17:50,759 --> 00:17:53,039 people study it for a really long time. Maybe the 368 00:17:53,279 --> 00:17:55,480 more fascinating piece is that we can actually use math 369 00:17:55,519 --> 00:17:55,920 to do that. 370 00:17:56,119 --> 00:17:59,160 Speaker 1: We're not discarding lessons learned of the past, actually figuring 371 00:17:59,200 --> 00:18:01,599 out how to use the physics and specifically in this case, 372 00:18:01,599 --> 00:18:05,279 the theoretical application in a real and real world scenario. 373 00:18:05,359 --> 00:18:06,880 You know, one of the things that I originally had 374 00:18:07,079 --> 00:18:09,640 thought of when we were talking about vector databases is that, like, 375 00:18:09,960 --> 00:18:13,880 surely this is only applicable to lms. But I feel 376 00:18:13,880 --> 00:18:16,440 like you said that the e commerce example that you 377 00:18:16,440 --> 00:18:19,000 brought up that really is about doing semantic search is 378 00:18:19,160 --> 00:18:22,319 like a simple example, And I don't know if. 379 00:18:22,240 --> 00:18:23,079 Speaker 2: I agree with you. 380 00:18:23,119 --> 00:18:25,759 Speaker 1: Actually, I feel like getting the search right and e 381 00:18:25,799 --> 00:18:29,119 commerce is like literally the most complicated example of search. 382 00:18:29,240 --> 00:18:31,920 Speaker 4: When I say simple, I mean it's simple for people 383 00:18:31,960 --> 00:18:37,279 to understand because as an example, because they shop online, right, 384 00:18:37,359 --> 00:18:39,039 everyone shop Most people shop online. 385 00:18:39,079 --> 00:18:39,839 Speaker 3: I don't want to say everyone. 386 00:18:39,880 --> 00:18:40,480 Speaker 2: I'm with you there. 387 00:18:40,759 --> 00:18:42,839 Speaker 1: I think there is something where it does seem simple 388 00:18:42,839 --> 00:18:44,640 in the service, but like as soon as you get 389 00:18:44,799 --> 00:18:47,640 to thinking about how the search actually works, it starts 390 00:18:47,640 --> 00:18:50,319 to get very complicated. I think one of the examples 391 00:18:50,319 --> 00:18:53,559 that comes up a lot for me when I've reviewed 392 00:18:53,680 --> 00:18:57,359 other companies database architecture, like which kinde of database they're 393 00:18:57,359 --> 00:18:59,680 going with, or you know what sort of indexes they have? 394 00:19:00,359 --> 00:19:02,160 It always sort as a generic problem, like oh, I 395 00:19:02,160 --> 00:19:03,519 have some data and I want to search it, and 396 00:19:03,519 --> 00:19:04,839 I'm like, well, what kind of search are you're doing. 397 00:19:04,880 --> 00:19:07,279 It's like, well, we have some attributes on each of 398 00:19:07,319 --> 00:19:10,079 the rows of our data or the items, and we 399 00:19:10,119 --> 00:19:12,400 want to filter by the attributes. And I'm like, well, 400 00:19:12,400 --> 00:19:13,960 which attributes do you want to filter by? Is it 401 00:19:13,960 --> 00:19:16,920 always like each item has three attributes and you always 402 00:19:16,920 --> 00:19:19,440 filter that on the first attribute and the second, the 403 00:19:19,640 --> 00:19:21,880 second and third are never used, or is it something 404 00:19:21,920 --> 00:19:25,039 complin or something simple? And like, oh, we know it's 405 00:19:25,079 --> 00:19:28,640 always well, the user could give us attribute one, two, 406 00:19:28,839 --> 00:19:30,920 or three, and we need to figure out all of 407 00:19:30,920 --> 00:19:33,079 the items that match one, two or three. And I'm like, well, 408 00:19:33,079 --> 00:19:36,160 that pretty much just eliminated every single note SQL database 409 00:19:36,200 --> 00:19:38,839 out there, because you know, good luck. I mean, I 410 00:19:38,880 --> 00:19:41,119 will say, in some scenarios you can be very clever 411 00:19:41,799 --> 00:19:46,240 where you can just index all three attributes, make three 412 00:19:46,279 --> 00:19:48,680 quaries to your database, and then merge the results. 413 00:19:48,759 --> 00:19:50,839 Speaker 4: I'm just gonna say, but also your products might all 414 00:19:50,880 --> 00:19:52,640 have different number of attributes. 415 00:19:52,720 --> 00:19:55,119 Speaker 2: Oh yeah, oh yeah, for sure, right. 416 00:19:55,519 --> 00:19:57,599 Speaker 1: And actually think the one that that's more complicated is 417 00:19:57,640 --> 00:19:59,480 that like some of the attributes have like an array 418 00:19:59,559 --> 00:20:04,279 of values, right, so it's like, oh, yeah, well this 419 00:20:04,359 --> 00:20:07,160 product we use these commerce example, like this product is 420 00:20:07,200 --> 00:20:09,200 a shirt. It can be in small, medium, and large, 421 00:20:09,200 --> 00:20:11,279 like the sizes that are available. And so when someone 422 00:20:11,359 --> 00:20:13,559 search sizes, you're not going to have an attribute column 423 00:20:13,599 --> 00:20:16,640 that's like you know, small exists, you know, true or 424 00:20:16,640 --> 00:20:19,119 falls you know, you know, shirt can be in medium 425 00:20:19,359 --> 00:20:19,960 true or falls. 426 00:20:19,960 --> 00:20:22,559 Speaker 2: Like this is nonsensical. So I think once you. 427 00:20:22,559 --> 00:20:25,000 Speaker 1: See that, you're like, well, okay, search and e commerce, 428 00:20:25,039 --> 00:20:25,920 that's that's. 429 00:20:25,759 --> 00:20:27,880 Speaker 2: Got to be like the most complicated thing ever. 430 00:20:27,920 --> 00:20:31,039 Speaker 1: And to say like we maybe collectively we may have 431 00:20:31,039 --> 00:20:33,200 stumbled upon the the. 432 00:20:33,200 --> 00:20:34,839 Speaker 2: Limit for improving search. 433 00:20:34,880 --> 00:20:37,839 Speaker 1: In that way, by using an embedding model, we're taking 434 00:20:37,880 --> 00:20:40,839 the core, the essence of what the search query is, 435 00:20:41,039 --> 00:20:42,680 trying to figure out what it is, and then mapping 436 00:20:42,720 --> 00:20:44,960 it to an optimized way of storying the data and 437 00:20:45,000 --> 00:20:47,519 a database that isn't a role based where you're doing 438 00:20:48,000 --> 00:20:51,960 column by column matching or some sort of non efficient 439 00:20:52,000 --> 00:20:53,920 index look up, like in a no SQL way where 440 00:20:53,960 --> 00:20:57,039 you're just like somehow you know exactly which item the 441 00:20:57,200 --> 00:20:58,680 person is talking about. 442 00:20:58,680 --> 00:21:00,480 Speaker 2: But I think that'd be pretty great. Reality you get. 443 00:21:00,359 --> 00:21:03,359 Speaker 1: To where I just, you know, search for red shirt 444 00:21:03,599 --> 00:21:07,839 and I get exactly the green trousers that I actually wanted. 445 00:21:07,920 --> 00:21:09,920 Speaker 4: I don't know if you're familiar with like the the 446 00:21:10,000 --> 00:21:14,759 dupe trend, the duplicates of like high end fashion, where 447 00:21:14,799 --> 00:21:18,559 these like low, lower end, cheaper alternatives come up with dupes. 448 00:21:19,440 --> 00:21:21,799 So we see it a lot in fashion, It's it's 449 00:21:21,839 --> 00:21:25,519 in other other areas too, But what if you could 450 00:21:25,519 --> 00:21:29,559 search for like this, this high end fashion brand name 451 00:21:30,039 --> 00:21:34,000 paired with like white t shirt with rough frilly edges 452 00:21:34,119 --> 00:21:37,720 of the sleeves, and then you come up with like 453 00:21:37,799 --> 00:21:40,759 a different the actual duplicate of that. 454 00:21:41,240 --> 00:21:42,799 Speaker 3: On this particular site. 455 00:21:43,039 --> 00:21:46,079 Speaker 4: I agree that like these products that you're you're talking 456 00:21:46,119 --> 00:21:48,720 about in the categories and in the different ways that 457 00:21:48,880 --> 00:21:52,079 it's not even just categories but just the different features 458 00:21:52,079 --> 00:21:55,799 of them are very complicated. But what if we could 459 00:21:55,799 --> 00:21:59,400 actually search for those based off of, you know, a 460 00:21:59,440 --> 00:22:02,119 few words that describe what it is, but don't necessarily 461 00:22:02,200 --> 00:22:03,799 use those same words you brought up. 462 00:22:03,960 --> 00:22:07,400 Speaker 1: I think one of the major problems with giant online 463 00:22:07,440 --> 00:22:11,440 search engines that are dedicated to consumer buying experience. And 464 00:22:11,480 --> 00:22:13,880 I'll say I think there's one that comes to mind 465 00:22:14,079 --> 00:22:15,559 in the Western. 466 00:22:15,160 --> 00:22:17,279 Speaker 2: World, which is Amazon. They have a. 467 00:22:17,279 --> 00:22:20,160 Speaker 1: Huge problem with basically I don't know if they call 468 00:22:20,200 --> 00:22:24,039 them duplicates as much as counterfeits, where if you send 469 00:22:24,240 --> 00:22:27,559 if you're not doing the manufacturing yourself, that you've sent 470 00:22:27,599 --> 00:22:30,200 the pattern to another company, and in their off time 471 00:22:30,200 --> 00:22:33,039 when they have extra resources, they just print out more 472 00:22:33,160 --> 00:22:36,559 of your item without the logo on it or with 473 00:22:36,599 --> 00:22:39,359 the logo because they don't care. And then when you 474 00:22:39,400 --> 00:22:42,359 go and search on any website you for the thing 475 00:22:42,400 --> 00:22:44,559 you're looking for, even with the brand name, you're getting 476 00:22:44,759 --> 00:22:48,720 the competitor version that's just not as good and people 477 00:22:48,720 --> 00:22:50,440 can't tell the difference most of the time, and this 478 00:22:50,480 --> 00:22:53,200 actually destroys a lot of small businesses that are doing it. 479 00:22:54,079 --> 00:22:57,240 I'm wondering if there really is a smart strategy here 480 00:22:57,279 --> 00:23:00,640 where you can use this for fraud detection. And I 481 00:23:00,680 --> 00:23:02,559 know when I'm searching on Amazon, I always want like 482 00:23:03,200 --> 00:23:06,480 to filter by results where the brand of the product 483 00:23:06,559 --> 00:23:09,599 matches the store where the product is coming from. Like that, 484 00:23:09,599 --> 00:23:12,319 that's almost always what I want. You make your thing 485 00:23:12,519 --> 00:23:14,200 that I don't care what your brand is, but it 486 00:23:14,640 --> 00:23:17,319 should match, and I always get suspicious when it's when 487 00:23:17,319 --> 00:23:23,279 it's something else. Yeah, yeah, I can see who who 488 00:23:23,359 --> 00:23:25,160 uses Amazon more than most? 489 00:23:25,240 --> 00:23:25,799 Speaker 2: Yeah for sure. 490 00:23:26,400 --> 00:23:28,000 Speaker 1: But I mean there is an interesting thing here, like 491 00:23:28,079 --> 00:23:31,519 do you think that not just for search generically for 492 00:23:31,720 --> 00:23:35,640 end users? Are there primary applications for a vector databased 493 00:23:35,640 --> 00:23:38,079 outside of search that you see as like this is 494 00:23:38,240 --> 00:23:41,599 just now a new strategy where if you're not using 495 00:23:41,880 --> 00:23:44,480 pine Cone or you know, one of the competitors to 496 00:23:44,519 --> 00:23:46,279 do this, you're really missing out on some of the 497 00:23:46,279 --> 00:23:48,000 core value that could be being provided. 498 00:23:48,119 --> 00:23:50,039 Speaker 4: I think you touch on kind of a hard problem. 499 00:23:50,279 --> 00:23:52,839 You mentioned fraud. I know, like I don't. 500 00:23:52,960 --> 00:23:55,240 Speaker 3: This is an area I don't know very much about. 501 00:23:55,000 --> 00:23:57,880 Speaker 4: But I know that it is a use case that 502 00:23:57,920 --> 00:24:01,720 we have seen people use vector data bases for to 503 00:24:01,880 --> 00:24:07,039 find uh, identifying patterns and of of fraudulent use in 504 00:24:07,079 --> 00:24:07,680 that type of thing. 505 00:24:07,880 --> 00:24:09,880 Speaker 3: So I can see that like that that could be 506 00:24:09,960 --> 00:24:10,319 a thing. 507 00:24:10,519 --> 00:24:14,480 Speaker 1: Do you know what sort of customers are primarily the 508 00:24:14,640 --> 00:24:17,359 pine Cone is looking for, like or you know, the 509 00:24:17,359 --> 00:24:20,599 the ideal customer profile that usually makes a good match, 510 00:24:20,640 --> 00:24:22,200 Like do you like not pay too much attention to 511 00:24:22,240 --> 00:24:25,119 like the specific use case and it's uh company segment 512 00:24:25,200 --> 00:24:27,480 or vertical or is it you see something about their 513 00:24:27,480 --> 00:24:30,279 technology stack that is a is a good match for you. 514 00:24:30,519 --> 00:24:36,319 Speaker 4: I'm working with developers to help them learn and understand 515 00:24:36,480 --> 00:24:39,839 how to use pine Cone, how to use a vector database, 516 00:24:39,880 --> 00:24:42,960 how to incorporate retrieval into their systems. But I would 517 00:24:43,039 --> 00:24:45,039 say that like one of the things we look for 518 00:24:45,240 --> 00:24:50,960 are people with our companies with like large quantities. 519 00:24:50,480 --> 00:24:51,920 Speaker 3: Of high dimensional data. 520 00:24:51,960 --> 00:24:56,839 Speaker 4: So this is going to be your emails, your your contracts, 521 00:24:56,920 --> 00:25:02,480 your PDF documents in is potentially video or audio, and 522 00:25:02,799 --> 00:25:06,359 need to get insight from it, whether that is you know, 523 00:25:06,799 --> 00:25:09,720 a search results as far as like the e commerce 524 00:25:09,720 --> 00:25:14,519 example we've been using, or it is insights about a 525 00:25:14,559 --> 00:25:17,839 particular you know, business unit and like how they are 526 00:25:17,920 --> 00:25:21,200 operating or how they are how they function. Right before this, 527 00:25:22,000 --> 00:25:24,799 I was reading one of our case studies about a company. 528 00:25:24,920 --> 00:25:29,920 It's about a medical company that's doing research on medications 529 00:25:30,039 --> 00:25:34,839 and they are searching over molecules Like that's a lot, right, 530 00:25:34,920 --> 00:25:37,960 and so like they are essentially embedding those molecules as 531 00:25:38,079 --> 00:25:40,720 vectors and then doing searches over those in order to 532 00:25:41,200 --> 00:25:44,640 gain insights and do research on their their work in 533 00:25:44,759 --> 00:25:46,240 order to develop medicine. 534 00:25:46,279 --> 00:25:47,119 Speaker 3: So I thought, I thought. 535 00:25:47,000 --> 00:25:49,319 Speaker 4: That was really interesting, just in that it's not just 536 00:25:49,759 --> 00:25:51,640 the e commerce examples. 537 00:25:51,799 --> 00:25:53,559 Speaker 2: I wanted to ask because you're on the other side. 538 00:25:53,839 --> 00:25:56,799 Speaker 1: I just like whether or not the ICP of the 539 00:25:56,920 --> 00:25:59,839 of the potential customers, like it does match the types 540 00:25:59,839 --> 00:26:02,680 of of questions or challenges the engineers who come to 541 00:26:02,759 --> 00:26:06,480 community workspaces to like specifically ask questions about like whether 542 00:26:06,559 --> 00:26:08,200 or not they're like already going in the right direction 543 00:26:08,440 --> 00:26:11,119 and the sorts of product areas they're focused on, is 544 00:26:11,119 --> 00:26:13,000 a good match for that, Or you see people just 545 00:26:13,119 --> 00:26:15,640 trying to use vector databases in places that like have 546 00:26:15,799 --> 00:26:18,200 no reason to be used there whatsoever. 547 00:26:18,480 --> 00:26:20,039 Speaker 3: There's probably a mix of both. 548 00:26:20,279 --> 00:26:24,119 Speaker 4: I see people who are they've heard of a vector database, 549 00:26:24,160 --> 00:26:27,599 they've heard of pine Cone, and they like, they're. 550 00:26:27,400 --> 00:26:28,359 Speaker 3: Like me, They're a developer. 551 00:26:28,440 --> 00:26:31,119 Speaker 4: They like knowing, like what's the new shiny thing, and 552 00:26:31,160 --> 00:26:32,480 so they want to learn about it. They want to 553 00:26:32,519 --> 00:26:35,319 learn how it might solve their problem. More ideal customers 554 00:26:35,359 --> 00:26:37,880 are people who have done are a little bit further 555 00:26:37,920 --> 00:26:40,039 in their journey, and so they understand, like what is 556 00:26:40,079 --> 00:26:41,799 the purpose of it, what are some of the problems 557 00:26:41,839 --> 00:26:44,519 that they can solve? They understand that, like their their 558 00:26:44,559 --> 00:26:49,039 problem fits in in some way, and we have a 559 00:26:49,079 --> 00:26:51,519 team that helps them figure out how it fits in 560 00:26:51,599 --> 00:26:56,200 and and how how to actually implement it at production scale. 561 00:26:56,279 --> 00:26:59,680 Speaker 1: Do you see them coming over from first the technical problem, 562 00:27:00,039 --> 00:27:04,240 then realizing they need a vector database to store their 563 00:27:04,359 --> 00:27:08,240 semantic embeddings and then go to pine Cone or do 564 00:27:08,279 --> 00:27:11,119 you see it as a sort of a nuanced play 565 00:27:11,279 --> 00:27:14,440 on top of generic databases or event open source databases 566 00:27:14,480 --> 00:27:17,880 that do offer a vector database, And it's like, well, 567 00:27:17,920 --> 00:27:19,759 you know, if you're doing something in the space, you 568 00:27:19,799 --> 00:27:24,200 may be fine, but if you need something more robust 569 00:27:24,359 --> 00:27:26,960 or at scale, like you would want to switch in 570 00:27:27,000 --> 00:27:27,279 a way? 571 00:27:27,319 --> 00:27:29,680 Speaker 4: Do you the developers that I've been interacting with, I 572 00:27:29,720 --> 00:27:33,200 think they are first and foremost interested in a piece 573 00:27:33,240 --> 00:27:37,319 of technology. They know that they have some data they 574 00:27:37,319 --> 00:27:40,519 want to make their retrieval augmented generation pipeline more accurate, 575 00:27:40,640 --> 00:27:43,319 so they understand that retrieval is a piece of that. 576 00:27:43,559 --> 00:27:46,519 And this is that's all I really know about about 577 00:27:46,559 --> 00:27:50,400 it so far. There's definitely other teams that I work 578 00:27:50,480 --> 00:27:52,880 with that are are definitely more in the weeds with 579 00:27:52,960 --> 00:27:55,279 people and like where they are in their journey as 580 00:27:55,319 --> 00:27:57,839 far as like using a vector database or using coming 581 00:27:57,839 --> 00:28:00,000 to pine Cone specifically. 582 00:27:59,519 --> 00:28:01,720 Speaker 1: Just go and use pine Cone, don't don't use a 583 00:28:02,079 --> 00:28:06,000 There's no reason using generic other vector database, especially, I mean, 584 00:28:06,119 --> 00:28:09,799 honestly from what I the research I've done, that trying 585 00:28:09,799 --> 00:28:12,519 to build your own model to get the embeddings working 586 00:28:12,640 --> 00:28:15,160 right is just such a huge lift in the first place. 587 00:28:15,200 --> 00:28:19,799 And given the challenges from ensuring like similar like you 588 00:28:19,839 --> 00:28:23,480 can't just switch your model from one version. Maybe maybe 589 00:28:23,519 --> 00:28:25,000 you're just going to tell me I'm totally wrong here. 590 00:28:25,200 --> 00:28:28,079 Don't upgrade your model without also replacing all of your embeddings, 591 00:28:28,119 --> 00:28:30,000 because the new results won't make any sense. 592 00:28:29,759 --> 00:28:31,440 Speaker 4: If you want to swap out your embeddings, Like if 593 00:28:31,440 --> 00:28:34,599 you've you've done evaluations, you've figured out, you've done testing, 594 00:28:34,640 --> 00:28:37,200 you've figured out that like it's no longer accurate enough, 595 00:28:37,400 --> 00:28:39,079 you're going to have to reembed that data using a 596 00:28:39,079 --> 00:28:41,720 different model, And so there are approaches to doing some 597 00:28:41,799 --> 00:28:44,680 benchmarking and testing to make sure that like your accuracy 598 00:28:44,799 --> 00:28:48,839 is in the right the acceptable frame of use case 599 00:28:48,880 --> 00:28:52,440 can actually support you would be swapping out the model 600 00:28:52,480 --> 00:28:53,680 and re embedding that data. 601 00:28:53,839 --> 00:28:55,359 Speaker 1: That makes sense to me, But that means that there's 602 00:28:55,400 --> 00:28:58,599 a huge extra cost here to doing a model upgrade, 603 00:28:58,880 --> 00:29:01,799 not just on like future searches and whatever. Like even 604 00:29:01,799 --> 00:29:05,039 if the model is faster you're building it yourself, there's 605 00:29:05,039 --> 00:29:07,920 some improvement or using an open source one, there's some 606 00:29:08,000 --> 00:29:10,039 driver there, but that's going to come with a cost. 607 00:29:10,119 --> 00:29:12,319 And I can see that to be a huge reason 608 00:29:12,440 --> 00:29:14,880 to go with like just take that all off the 609 00:29:14,920 --> 00:29:17,119 table and if you know you need some sort of 610 00:29:17,119 --> 00:29:20,359 semantic search or some other strategy that uses embeddings, to 611 00:29:20,400 --> 00:29:23,680 go with a database that has that baked in automatically 612 00:29:23,960 --> 00:29:27,319 without having to think about how to do upgrades between models. 613 00:29:27,440 --> 00:29:31,000 Speaker 4: That's a big reason why we want to do evaluations 614 00:29:31,079 --> 00:29:34,519 and venuremarking ahead of time on potentially a smaller set 615 00:29:34,559 --> 00:29:39,000 of data before like committing is because you're right, it is. 616 00:29:39,240 --> 00:29:42,559 Speaker 3: It takes time, it costs money to do this. 617 00:29:42,839 --> 00:29:45,440 Speaker 4: Not everybody is going to have the time and money 618 00:29:45,480 --> 00:29:48,119 and expertise to find to or train their own model. 619 00:29:48,279 --> 00:29:51,319 I've mentioned retrival augmented generation a few times here. I've 620 00:29:51,359 --> 00:29:56,200 been spending essentially the last quarter I've been not only 621 00:29:56,200 --> 00:29:58,559 digging in myself and trying to learn about the different 622 00:29:59,079 --> 00:30:02,799 the different parts of but also different like approaches to 623 00:30:02,880 --> 00:30:06,519 doing it and sharing some of that. I've shared some 624 00:30:06,559 --> 00:30:10,079 of that publicly, I'm doing stuff internally related to that, 625 00:30:10,200 --> 00:30:13,400 and also in the future here I've got some stuff 626 00:30:13,400 --> 00:30:16,480 going on. But we are seeing people doing this and 627 00:30:16,519 --> 00:30:22,160 not fully understanding what they're doing. Shocker, yeah, right, exactly. 628 00:30:22,200 --> 00:30:26,079 Probably multiple reasons for that, Like we can we could 629 00:30:26,200 --> 00:30:29,279 go down the Vibe coding rabbit hole of like how 630 00:30:29,319 --> 00:30:32,319 that is contributing to some of some of the good 631 00:30:32,359 --> 00:30:35,880 and bad parts of this. Like obviously it's it's enabling 632 00:30:35,920 --> 00:30:39,160 people to to do more and to like get further 633 00:30:39,240 --> 00:30:44,920 into their problem, but also it potentially brings more challenges 634 00:30:45,640 --> 00:30:47,519 then they even know how to handle. 635 00:30:47,640 --> 00:30:49,000 Speaker 2: I really do have to dive into that. 636 00:30:49,039 --> 00:30:52,119 Speaker 1: So from a vibe coding standpoint, our vector database is 637 00:30:52,160 --> 00:30:56,359 being recommended by coding assistants as a solution for a 638 00:30:56,400 --> 00:30:57,759 problem like it was just like oh, yeah, you know, 639 00:30:57,839 --> 00:31:00,880 generating some code and it pulls in a way to 640 00:31:00,920 --> 00:31:04,079 write to an open source database that requires embeddings or 641 00:31:04,200 --> 00:31:05,480 is that just not happening yet. 642 00:31:05,559 --> 00:31:07,400 Speaker 3: I do know that it comes up. It does. 643 00:31:07,799 --> 00:31:11,759 Speaker 4: It does propose pine code and other competitors and stuff 644 00:31:11,799 --> 00:31:14,039 like that. One of the things that like we have 645 00:31:14,079 --> 00:31:17,039 a challenge with and I expect other people to do 646 00:31:17,079 --> 00:31:17,440 as well. 647 00:31:17,559 --> 00:31:19,759 Speaker 3: Is like, because these. 648 00:31:19,720 --> 00:31:22,319 Speaker 4: Models are trained on old data, like it's using our 649 00:31:22,400 --> 00:31:25,759 old data, our old document public documentation, and so it 650 00:31:25,799 --> 00:31:29,680 isn't always the most accurate. And so we we do 651 00:31:29,720 --> 00:31:35,200 stuff on our end to to try and encourage those 652 00:31:35,200 --> 00:31:38,839 models to or those tools, not necessarily the model itself, 653 00:31:38,880 --> 00:31:41,799 but the tools to you know, generate the right code, 654 00:31:41,839 --> 00:31:43,240 the most up to date code. 655 00:31:43,319 --> 00:31:45,319 Speaker 1: So we're still a little ways away from it being 656 00:31:45,359 --> 00:31:48,640 always the right answer popping up in lms. How about 657 00:31:48,680 --> 00:31:52,240 the how about the LM companies, So companies that claim 658 00:31:52,279 --> 00:31:54,599 they have some sort of AI and they are really 659 00:31:54,680 --> 00:31:57,079 just you know, an LM that's solving a particular use case. 660 00:31:57,200 --> 00:32:02,359 Are they the cornerstone for company case where like a 661 00:32:02,400 --> 00:32:06,039 case study that would be using RAG more often than 662 00:32:06,079 --> 00:32:09,759 not there or is it just there's a spectrum and 663 00:32:09,799 --> 00:32:13,559 it really depends on the vertical or market segment or 664 00:32:13,680 --> 00:32:14,440 product area. 665 00:32:15,200 --> 00:32:18,839 Speaker 4: I mean there's probably a lot of opinions there. I 666 00:32:18,839 --> 00:32:22,440 think that's like depending on who you talk to on 667 00:32:22,920 --> 00:32:26,160 how that is. I think like for me, I see 668 00:32:26,920 --> 00:32:31,599 maybe the model companies are not necessarily advocating for retrieval 669 00:32:31,640 --> 00:32:34,240 augmented generation I mean maybe they are, but like we 670 00:32:34,319 --> 00:32:36,880 keep seeing these models get bigger and better and faster, right, 671 00:32:37,759 --> 00:32:41,000 but there's still those limitations that I mentioned from the beginning, right, 672 00:32:41,319 --> 00:32:43,960 like they're only trained up until a certain period of time. 673 00:32:44,039 --> 00:32:44,920 Speaker 3: There's only. 674 00:32:46,160 --> 00:32:48,200 Speaker 4: It doesn't it's not trained on your company data, your 675 00:32:48,240 --> 00:32:52,440 your private data, and so among a few other limitations. 676 00:32:52,440 --> 00:32:54,839 But those are kind of the key ones that people 677 00:32:54,920 --> 00:32:58,680 really recognize. And so that's where this retrieval part of 678 00:32:59,240 --> 00:33:01,480 RAG is coming in, is like it's the way to 679 00:33:01,640 --> 00:33:04,880 give your model more knowledge, to give it more accurate 680 00:33:04,880 --> 00:33:06,200 and authoritative knowledge. 681 00:33:06,319 --> 00:33:09,640 Speaker 1: So if the primary use case outside like semantic searching 682 00:33:09,839 --> 00:33:13,440 or some sort of comparison search, if you're utilizing it 683 00:33:13,480 --> 00:33:17,359 to extend your data set, either from your own data 684 00:33:17,559 --> 00:33:19,799 or something else that you want to pull in, and 685 00:33:19,839 --> 00:33:20,839 that means you're using RAG. 686 00:33:20,920 --> 00:33:22,440 Speaker 2: Than if you're using RAG, that means you're using a 687 00:33:22,480 --> 00:33:23,240 vector database. 688 00:33:23,400 --> 00:33:27,279 Speaker 3: Not always. Oh it's interesting, it could be retrieving from 689 00:33:27,319 --> 00:33:29,880 some other database. That is, if we think about it. 690 00:33:29,960 --> 00:33:33,200 Speaker 4: I think you recently did like an MCP episode, right, 691 00:33:33,599 --> 00:33:37,000 So that's one approach to getting to kind of interfacing 692 00:33:37,039 --> 00:33:41,079 with other tools and services or even like chat GPT. 693 00:33:41,279 --> 00:33:42,400 Speaker 3: It can go out to the internet. 694 00:33:42,400 --> 00:33:44,759 Speaker 4: That is a tool that it's using to go get 695 00:33:44,799 --> 00:33:48,240 other data and augment your result, your output with more 696 00:33:48,279 --> 00:33:49,119 accurate data. 697 00:33:49,200 --> 00:33:51,640 Speaker 1: One thing I noticed, especially with companies that are still 698 00:33:51,960 --> 00:33:55,759 incredibly technical but aren't specifically in any of the quote 699 00:33:55,799 --> 00:34:00,559 unquote AI spaces, is that they definitely get the difference 700 00:34:00,559 --> 00:34:01,839 between MCP. 701 00:34:01,880 --> 00:34:03,680 Speaker 2: And RAG wrong. 702 00:34:03,920 --> 00:34:07,319 Speaker 1: Often like the bijection of how many of one of 703 00:34:07,359 --> 00:34:09,719 these things they need versus another one? Like I was 704 00:34:09,800 --> 00:34:11,719 one of my colleagues is working out a very interesting 705 00:34:11,719 --> 00:34:15,480 company where they have a need to do RAG on 706 00:34:15,559 --> 00:34:18,760 behalf of their customers, So they're they're pulling in knowledge 707 00:34:18,760 --> 00:34:21,679 bases from their customers and they're trying to figure out 708 00:34:21,679 --> 00:34:25,320 how that interplays with MCP servers. And one thing that 709 00:34:25,719 --> 00:34:28,239 has been problematic is they don't want to own the data. 710 00:34:28,760 --> 00:34:30,760 But at the same time, a lot of them MCP 711 00:34:30,920 --> 00:34:34,159 providers don't have this concept of like multi tenancy, Like 712 00:34:34,199 --> 00:34:38,360 they understand you as a customer can only access your data, 713 00:34:38,679 --> 00:34:41,079 but they don't have a good concept of how to 714 00:34:41,519 --> 00:34:43,480 group or sequester parts of. 715 00:34:43,440 --> 00:34:46,159 Speaker 2: The data into smaller into smaller areas. 716 00:34:45,760 --> 00:34:48,880 Speaker 4: From a vector database perspective, like the way we've implemented 717 00:34:48,960 --> 00:34:53,519 multi tenancy is through name spaces. So if you think 718 00:34:53,639 --> 00:34:56,760 of a company that is offering agents to their customers, 719 00:34:56,800 --> 00:35:00,199 each agent or each user that they are that as 720 00:35:00,239 --> 00:35:02,880 an agent would potentially be within its own name space. 721 00:35:03,199 --> 00:35:06,840 Speaker 3: So that is a way to like actually segregate the data. 722 00:35:07,280 --> 00:35:10,119 Speaker 4: I think you touched on something interesting though, that like 723 00:35:10,679 --> 00:35:14,199 not all the companies out there are AI first companies 724 00:35:14,480 --> 00:35:17,599 or they are not like well versed in AI technologies 725 00:35:17,639 --> 00:35:20,639 and solutions. And I think we're at the point where 726 00:35:20,679 --> 00:35:23,119 that's going to become a huge thing because all of 727 00:35:23,159 --> 00:35:25,280 these companies, there's so many more of those companies out 728 00:35:25,280 --> 00:35:28,599 there that then there are AI companies that have been 729 00:35:28,599 --> 00:35:30,239 doing this for a long time, right. 730 00:35:30,599 --> 00:35:32,679 Speaker 1: I think we're going to have a fight on this episode, 731 00:35:32,840 --> 00:35:35,840 So you know, I'm going to quote some research that 732 00:35:35,880 --> 00:35:39,639 I think came out of Nonda AI report from MIT 733 00:35:40,000 --> 00:35:43,320 that said only five percent of companies are getting value 734 00:35:43,400 --> 00:35:46,119 out of implemented l Like I'm going to. 735 00:35:46,119 --> 00:35:48,280 Speaker 2: Say quote unquote AI because I don't think we have AI. 736 00:35:48,320 --> 00:35:50,000 Speaker 1: That's a different episode where I got into that, but 737 00:35:50,400 --> 00:35:51,880 we can say we can say AI for the rest 738 00:35:51,880 --> 00:35:55,440 of this one aren't getting the value out and it's 739 00:35:55,599 --> 00:36:00,119 just a huge cost SYNC time saying resourcenc. If you 740 00:36:00,159 --> 00:36:04,119 say that going forward, there's gonna it's gonna be interesting. 741 00:36:04,679 --> 00:36:07,320 I can either read that as you believe that companies 742 00:36:07,360 --> 00:36:12,440 will transition to having AI, or all the companies who 743 00:36:12,480 --> 00:36:14,559 don't do it will be out of business and therefore 744 00:36:14,559 --> 00:36:17,880 the only companies left will be ones who do AI. 745 00:36:18,280 --> 00:36:19,360 Speaker 3: I don't mean the latter. 746 00:36:19,639 --> 00:36:22,360 Speaker 4: I think there is going to be a transition, but 747 00:36:22,400 --> 00:36:25,320 I think there's an opportunity there. My background is not 748 00:36:25,480 --> 00:36:28,360 in AI, like me coming to pine Cone is that's 749 00:36:28,440 --> 00:36:30,679 this is a new space for me, right, So I 750 00:36:30,719 --> 00:36:34,719 bring the lens of the traditional developer who's been tasked 751 00:36:34,719 --> 00:36:37,960 with a problem. And in this case, the problem these 752 00:36:38,039 --> 00:36:41,440 days might be, you know, a problem that requires a 753 00:36:41,519 --> 00:36:44,480 vector database. And so as someone who has been a 754 00:36:44,480 --> 00:36:46,360 full stack developer, now I have to go out and 755 00:36:46,400 --> 00:36:48,360 figure out, like how do I solve this problem with 756 00:36:48,440 --> 00:36:50,239 this tool that I've been told to use. 757 00:36:50,599 --> 00:36:51,519 Speaker 3: For sure, I. 758 00:36:51,440 --> 00:36:53,000 Speaker 4: Think we're going to see more and more of that. 759 00:36:53,320 --> 00:36:55,559 Definitely don't mean that companies are going to go away. 760 00:36:55,760 --> 00:36:58,840 Like obviously, like people are still running on mainframes, right 761 00:36:58,880 --> 00:37:00,840 and so like it might not fit in that space, 762 00:37:01,039 --> 00:37:04,000 technology sticks around for a really long time. I think 763 00:37:04,039 --> 00:37:07,679 there's going to be some change and it's not always 764 00:37:07,679 --> 00:37:08,599 going to be easy. 765 00:37:08,840 --> 00:37:11,760 Speaker 1: And I wonder what the turnaround is for this, because 766 00:37:12,360 --> 00:37:17,239 I see companies still claiming that they're introducing Agile and 767 00:37:18,239 --> 00:37:20,159 like that, or worse, they like they've done it. 768 00:37:20,199 --> 00:37:21,519 Speaker 3: I thought Agile was gone. 769 00:37:21,800 --> 00:37:24,760 Speaker 1: Well, I mean, okay, so the manifesto was like in 770 00:37:24,800 --> 00:37:28,119 two thousand or like two thousand and one or something. 771 00:37:28,159 --> 00:37:31,840 I'm terrible with years. I knew what happened before I 772 00:37:31,920 --> 00:37:36,199 really got into software engineering. I think realistically, either companies 773 00:37:36,239 --> 00:37:39,400 say they do it and they don't, or they acknowledge 774 00:37:39,400 --> 00:37:41,000 that they don't do it, and that for me is 775 00:37:41,039 --> 00:37:44,199 like mind boggling, because from my standpoint, everyone should be 776 00:37:44,199 --> 00:37:47,199 doing it all the time. I can appreciate the belief 777 00:37:47,280 --> 00:37:49,639 that that the same thing will happen with AI. But 778 00:37:49,840 --> 00:37:52,639 if you know we're twenty four years out, twenty five 779 00:37:52,719 --> 00:37:57,440 years out from the Agile Manifesto, then I think we're 780 00:37:57,719 --> 00:37:58,639 we're forever away. 781 00:38:01,159 --> 00:38:04,119 Speaker 4: I don't necessarily disagree with you. I think like there's 782 00:38:04,159 --> 00:38:08,480 a lot of legacy code and applic I mentioned main mainframe, right, Yeah, 783 00:38:08,519 --> 00:38:10,519 there are a lot of different companies who are just 784 00:38:10,559 --> 00:38:13,760 at different stages of their maturity in their organization. Some 785 00:38:13,840 --> 00:38:16,400 are big, some are small, and not all of them 786 00:38:16,519 --> 00:38:18,760 are going to get there at the same time. You're 787 00:38:18,760 --> 00:38:21,320 probably familiar with it, but there's this curve of like 788 00:38:21,400 --> 00:38:24,400 where people are at on the the acceptance of a 789 00:38:24,440 --> 00:38:25,719 particular product. 790 00:38:25,440 --> 00:38:26,920 Speaker 2: Or they're crossing the chasm. 791 00:38:27,079 --> 00:38:28,920 Speaker 3: Yes, yes, exactly, that's what it is. 792 00:38:29,079 --> 00:38:32,199 Speaker 4: One thing that I think we have seen is people 793 00:38:32,280 --> 00:38:38,239 are running more production workloads now with pine Cone, whereas 794 00:38:38,239 --> 00:38:40,920 in the past it has like it has been I 795 00:38:40,920 --> 00:38:42,800 don't want to say it has been less production, but 796 00:38:43,519 --> 00:38:45,639 we are seeing more people kind of latch onto that 797 00:38:45,719 --> 00:38:49,440 and doing stuff in production. And so as as we 798 00:38:49,480 --> 00:38:52,280 see more and more of that, like that's that's people 799 00:38:52,320 --> 00:38:55,239 like learning how to do this. So like like I'm saying, 800 00:38:55,280 --> 00:38:57,039 like we're still very early in. 801 00:38:56,960 --> 00:38:59,159 Speaker 1: This, I'm totally on the same page there, and I 802 00:38:59,400 --> 00:39:02,440 think absolutely right those workloads may or may not have 803 00:39:02,440 --> 00:39:05,400 anything to do with LMS. Like we've solved We've identified 804 00:39:05,400 --> 00:39:08,880 a new functional way of storing data and searching it, 805 00:39:09,239 --> 00:39:12,280 and there are primary applications where that's valuable. If you 806 00:39:12,320 --> 00:39:14,679 talk about RAG, we're talking about knowledge bases, we're talking 807 00:39:14,719 --> 00:39:17,119 about semantic search, where you know, we're talking about searching 808 00:39:17,159 --> 00:39:20,440 through tons of attributes and fuzzy searching, free tech searching 809 00:39:20,480 --> 00:39:23,599 that never really was that great to begin with. I mean, 810 00:39:23,719 --> 00:39:27,440 whole companies have made their entire existence to figure out 811 00:39:27,480 --> 00:39:29,639 how to actually do this correctly. And now we have 812 00:39:29,679 --> 00:39:32,719 a fundamental technology or understanding of how to do this 813 00:39:32,800 --> 00:39:37,639 fundamentally at a principle level with you know, raw mathematics. 814 00:39:37,880 --> 00:39:40,159 I do think that there is some coming around there, 815 00:39:40,159 --> 00:39:42,400 and you know, it will be interesting to see whether 816 00:39:42,480 --> 00:39:45,840 or not the companies that gravitate towards needing vector databases 817 00:39:45,880 --> 00:39:48,159 are more and more in the alarm space or less 818 00:39:48,159 --> 00:39:51,199 and less as companies figure out the value that they 819 00:39:51,239 --> 00:39:55,199 can extract from it. So I guess it'll be interesting 820 00:39:55,199 --> 00:39:58,199 to see where where pine Cone's customers end up a 821 00:39:58,239 --> 00:40:00,760 few years from now. I do want to quit maybe 822 00:40:00,840 --> 00:40:03,559 ask you an opinion on something. I think one thing 823 00:40:03,559 --> 00:40:06,480 that has happened in the last five years is companies 824 00:40:06,599 --> 00:40:09,920 having to deal with LM's being used during the interview process. 825 00:40:10,480 --> 00:40:12,639 And I don't know if I should just stop the 826 00:40:12,719 --> 00:40:15,000 question there and just let you say something, or if 827 00:40:15,039 --> 00:40:19,559 I should specifically ask, like I just ask, yeah, has 828 00:40:19,639 --> 00:40:24,599 this impacted pine Cones interviewing practices and if yes or no, 829 00:40:24,719 --> 00:40:29,440 like are they seeing the use of llms intentionally my candidates, 830 00:40:29,440 --> 00:40:32,960 Like is it encouraged to be done or have you 831 00:40:33,000 --> 00:40:35,079 worked around trying to figure out how to deal with 832 00:40:35,119 --> 00:40:37,760 the fact that these lms will be used during the 833 00:40:38,119 --> 00:40:39,239 interviewing cycle. 834 00:40:39,159 --> 00:40:42,159 Speaker 4: In my experience, and I and at my last job, 835 00:40:42,199 --> 00:40:44,920 I did do a lot of also interviewing to hire, 836 00:40:46,280 --> 00:40:51,079 and it's definitely a thing people use them. I did 837 00:40:51,119 --> 00:40:53,760 not use them. It was not part of my interview process. 838 00:40:53,840 --> 00:40:57,880 But I do think as long as like you're setting 839 00:40:57,920 --> 00:41:00,880 the expectation of how and why you're using a tool, 840 00:41:01,960 --> 00:41:02,400 then it. 841 00:41:02,360 --> 00:41:03,159 Speaker 3: Makes more sense. 842 00:41:03,440 --> 00:41:07,599 Speaker 4: I don't think that anyone should be like pretending they 843 00:41:07,599 --> 00:41:08,679 know something when they don't. 844 00:41:08,920 --> 00:41:11,400 Speaker 1: You just destroyed the whole industry there, like just that 845 00:41:11,519 --> 00:41:12,159 one statement. 846 00:41:12,880 --> 00:41:15,920 Speaker 4: Yeah, yeah, I think like I think it just from 847 00:41:15,920 --> 00:41:18,679 a practical perspective, like we use tools in our day 848 00:41:18,760 --> 00:41:22,079 jobs and so like if I can't use cursor or 849 00:41:22,159 --> 00:41:26,159 cloud code to write code and show how I would 850 00:41:26,280 --> 00:41:29,639 use do my actual job, then you know, it's really 851 00:41:29,639 --> 00:41:33,039 hard for an interviewer to understand how you're going. 852 00:41:32,920 --> 00:41:33,599 Speaker 3: To do your job. 853 00:41:33,880 --> 00:41:36,360 Speaker 1: Interesting personal opinion, because and I think this is a 854 00:41:36,400 --> 00:41:38,320 pattern for a lot of our guests that come on 855 00:41:38,360 --> 00:41:43,760 this show that previously you were working at Amazon. I 856 00:41:43,760 --> 00:41:47,000 don't know if it was AWS specifically, so I don't 857 00:41:47,000 --> 00:41:49,000 know what it is with our podcast. And like people 858 00:41:49,039 --> 00:41:52,280 like leave AWS after being there for like four years 859 00:41:52,360 --> 00:41:53,639 or so, I didn't actually look to see how long 860 00:41:53,679 --> 00:41:56,000 you were there and then then are a new job 861 00:41:56,079 --> 00:41:58,639 for a couple of years and then come on with 862 00:41:58,760 --> 00:42:01,000 podcast and have some very interesting things to say. 863 00:42:01,320 --> 00:42:06,320 Speaker 3: So no spilling of the detail. Just on Earth five years. 864 00:42:07,039 --> 00:42:10,559 Speaker 1: There was a question, especially around the interviewing. Did you 865 00:42:10,559 --> 00:42:13,440 see things already starting to be rolled out at AWS. 866 00:42:12,880 --> 00:42:15,559 Speaker 4: During my almost five years I was interviewing people, and 867 00:42:15,599 --> 00:42:18,159 like it's a very rigorous, like you go through training 868 00:42:18,159 --> 00:42:20,840 to learn how to interview, and you've probably seen some 869 00:42:20,880 --> 00:42:23,840 of the questions that were we ask While I was there, 870 00:42:23,920 --> 00:42:26,599 there was no conversation, like I didn't have any conversations 871 00:42:26,639 --> 00:42:31,440 with my leadership about when or how lms are allowed 872 00:42:31,440 --> 00:42:34,400 to be used during the process. Like I don't want 873 00:42:34,400 --> 00:42:36,000 to say it was early enough where it wasn't happening. 874 00:42:36,000 --> 00:42:38,559 I'm sure it was happening, but it was not a 875 00:42:38,559 --> 00:42:40,000 part of my experience there. 876 00:42:40,079 --> 00:42:42,199 Speaker 2: During the interviewing I don't want to know. 877 00:42:42,280 --> 00:42:44,800 Speaker 3: I think it's the Yeah, I know right. 878 00:42:44,960 --> 00:42:46,000 Speaker 2: I'll say this, if. 879 00:42:45,920 --> 00:42:47,760 Speaker 1: You get through the interview round with us, you were 880 00:42:47,840 --> 00:42:50,039 using an LM, and then you continue to use an 881 00:42:50,119 --> 00:42:52,400 LM in your day job for however long you're here, 882 00:42:52,480 --> 00:42:54,960 until the day you leave, and no one finds out 883 00:42:55,159 --> 00:42:56,039 was there any harm? 884 00:42:57,199 --> 00:43:00,880 Speaker 4: For me, it's been like using an LM as part 885 00:43:00,920 --> 00:43:03,800 of my job has been embedded since she at GPT 886 00:43:03,960 --> 00:43:08,079 came out like interesting at Amazon, not just AWS, but 887 00:43:08,119 --> 00:43:11,039 at Amazon, like we were tasked with figure out how 888 00:43:11,039 --> 00:43:14,960 to use this yeah, and same thing at Pinecone, like 889 00:43:15,000 --> 00:43:17,559 we're we're tasked with figuring out how to use these 890 00:43:17,559 --> 00:43:20,320 tools because everyone else is using them. And if they're 891 00:43:20,320 --> 00:43:22,159 trying to use our products or they're trying to do 892 00:43:22,239 --> 00:43:26,079 stuff with pine Cone, then we need to know their experiences. 893 00:43:26,119 --> 00:43:28,719 We need to know what the friction is, what's working, 894 00:43:28,760 --> 00:43:30,920 what's not working, all that kind of stuff. When I say, 895 00:43:31,000 --> 00:43:33,719 I think it is okay as long as to use 896 00:43:33,719 --> 00:43:36,360 it during the interview process, as long as the expectations 897 00:43:36,400 --> 00:43:37,880 are the same on both sides. 898 00:43:38,199 --> 00:43:40,079 Speaker 3: It's because I use them every day. 899 00:43:40,280 --> 00:43:43,199 Speaker 4: I fully recognize like I'm in a special situation. I 900 00:43:43,239 --> 00:43:45,280 work for a company that deals with this stuff right 901 00:43:46,000 --> 00:43:46,880 and builds this stuff. 902 00:43:47,159 --> 00:43:48,760 Speaker 1: So this is actually why I like asking the question 903 00:43:48,800 --> 00:43:51,679 specifically from people that are working out a company that's 904 00:43:51,679 --> 00:43:55,280 in and around the space where they're building tools or 905 00:43:55,280 --> 00:43:57,760 support for things like it would be a weird twist 906 00:43:57,760 --> 00:43:59,519 of fate where you're like, well, we don't know how 907 00:43:59,519 --> 00:44:02,639 to deal with these these interviewers, Like these interviews where 908 00:44:02,639 --> 00:44:04,400 the candidate comes in as using an LLM and like 909 00:44:04,480 --> 00:44:07,000 the product that they're using is using some sort of 910 00:44:07,119 --> 00:44:10,280 RAG that has a vector database that's running on pine Cone. 911 00:44:10,320 --> 00:44:12,159 Like you know that the sort of like weird circle. 912 00:44:12,199 --> 00:44:14,960 There where probably something you should have thought about, but 913 00:44:15,039 --> 00:44:17,599 it's it's good to you know, realize. Okay, no, actually, 914 00:44:17,840 --> 00:44:21,239 not only are our employees but also are you know, 915 00:44:21,280 --> 00:44:23,840 the developers that are utilizing the product to embed in 916 00:44:23,840 --> 00:44:27,400 their own applications. They're utilizing these tools as an important 917 00:44:27,400 --> 00:44:31,840 part of the flow, like expectedly, so understanding how they're 918 00:44:31,880 --> 00:44:34,039 utilizing them is important for us to even design a 919 00:44:34,079 --> 00:44:34,840 better application. 920 00:44:35,239 --> 00:44:36,360 Speaker 3: Yeah, exactly. 921 00:44:36,519 --> 00:44:38,880 Speaker 1: Maybe I'll give you the moment in case there's anything 922 00:44:38,880 --> 00:44:41,039 that we left out here that you're just like, I 923 00:44:41,480 --> 00:44:44,559 really need to share about this thing. Be it a 924 00:44:44,599 --> 00:44:47,480 major pine Cone release. The best feature that has that 925 00:44:47,480 --> 00:44:49,639 has ever come out is about to drop, you know, 926 00:44:49,679 --> 00:44:51,480 one week from now. I don't know what that is, 927 00:44:51,519 --> 00:44:53,079 but you know it, feel free to plug that I 928 00:44:53,079 --> 00:44:53,480 don't have. 929 00:44:53,639 --> 00:44:56,000 Speaker 4: I don't have the next feature that's coming out, and 930 00:44:56,039 --> 00:44:58,800 if I did, I probably couldn't say so. It's a 931 00:44:58,840 --> 00:45:02,400 cool technology. So if you are a person who likes 932 00:45:02,440 --> 00:45:06,960 to understand like how things work, and we have a 933 00:45:07,000 --> 00:45:09,880 lot of really good content on the pine Cone blog 934 00:45:11,039 --> 00:45:14,599 about just kind of search in general and like algorithms 935 00:45:14,599 --> 00:45:17,280 in general, and if you're if that's the thing you're into. 936 00:45:17,119 --> 00:45:19,400 Speaker 1: And whether elms are here to stay or not. I 937 00:45:19,440 --> 00:45:22,719 do think that there is a unique innovation that's happened 938 00:45:22,719 --> 00:45:25,719 here and outside of that, it's definitely worth learning. So 939 00:45:25,920 --> 00:45:28,719 if this weekend you are going to spend some time 940 00:45:28,800 --> 00:45:32,800 rewriting whatever your favorite JavaScript run time is in a 941 00:45:32,840 --> 00:45:35,320 new language like Rust or zig just because you can 942 00:45:35,840 --> 00:45:38,039 or an operator for Kubernetes, it sounds like, you know, 943 00:45:38,320 --> 00:45:41,400 a better use of your time would be potentially to 944 00:45:41,440 --> 00:45:43,880 invest in learning a vector databases and why not use 945 00:45:43,920 --> 00:45:44,519 pine Cone for that? 946 00:45:44,679 --> 00:45:46,639 Speaker 3: I agree with that. I think upskilling is in the 947 00:45:46,679 --> 00:45:47,239 long run, So. 948 00:45:47,519 --> 00:45:50,679 Speaker 1: With there we can switch over to doing picks. So 949 00:45:50,840 --> 00:45:53,119 my pick for this week is how do I put this? 950 00:45:53,440 --> 00:45:56,519 I found that I often plug things into my computers 951 00:45:56,519 --> 00:45:59,519 and unplug them over and over again, and I frequently 952 00:45:59,559 --> 00:46:01,679 get concer learned about the reliability of. 953 00:46:01,599 --> 00:46:03,079 Speaker 2: The USBC port. 954 00:46:03,119 --> 00:46:05,880 Speaker 1: Everything I have is USBC, especially like my ub key. 955 00:46:06,000 --> 00:46:08,199 So I'm a big, you know, physical past key user 956 00:46:08,239 --> 00:46:11,519 because I'm I think too much about security, probably more 957 00:46:11,559 --> 00:46:15,320 than I should, and so I spend a ton of 958 00:46:15,360 --> 00:46:19,119 money more than I probably should un buying magnetic USBC connectors. 959 00:46:19,119 --> 00:46:21,039 So you know, if you're watching the video, it's like 960 00:46:21,159 --> 00:46:23,400 this piece here is just like magnetic which just like 961 00:46:23,480 --> 00:46:25,320 goes together and you can. 962 00:46:25,199 --> 00:46:26,000 Speaker 2: Just walk around with this. 963 00:46:26,079 --> 00:46:29,360 Speaker 1: Here's like a ub key, my micro one, and honestly, 964 00:46:29,360 --> 00:46:30,920 it's just made it better. I plug it like all 965 00:46:30,920 --> 00:46:34,800 my all my my laptops, my cell phone, all my 966 00:46:34,800 --> 00:46:37,280 connectors that are sitting out have a magnetic thing on it, 967 00:46:37,559 --> 00:46:39,280 and it's just been it's just been great for the 968 00:46:39,360 --> 00:46:40,039 last few months. 969 00:46:40,039 --> 00:46:41,679 Speaker 2: Like I can't believe I waited so long to do this. 970 00:46:42,079 --> 00:46:44,559 Speaker 4: Is this magnetic piece like a protector, so it's a 971 00:46:44,599 --> 00:46:46,599 cover when you're not using it or I don't I 972 00:46:46,639 --> 00:46:47,400 don't understand. 973 00:46:47,599 --> 00:46:50,639 Speaker 1: So the piece is just a USBC to USBC connector, 974 00:46:50,760 --> 00:46:53,559 so it's just USBC and this is also USBC and 975 00:46:53,599 --> 00:46:57,559 this part is two pieces, so this gets connected. So realistically, 976 00:46:57,840 --> 00:46:59,760 I walk around with this just this side, which is 977 00:46:59,760 --> 00:47:01,920 not as we see. It's just like a magnetic piece 978 00:47:01,960 --> 00:47:05,039 and it just they just snap together and they're interchangeable. 979 00:47:05,119 --> 00:47:07,679 Speaker 2: So like the same one I use for my cell phone, 980 00:47:07,719 --> 00:47:08,679 like I could, I could. 981 00:47:08,760 --> 00:47:10,199 Speaker 1: I don't know, no one's gonna be able to see this, 982 00:47:10,239 --> 00:47:11,480 but if I hold up my cell phone with the 983 00:47:11,519 --> 00:47:13,280 piece and I take my ubiki, like it will snap 984 00:47:13,320 --> 00:47:16,159 onto here. But it's the same thing that also, like 985 00:47:16,199 --> 00:47:19,199 if I want to charge my phone, it's the same connector. 986 00:47:19,880 --> 00:47:22,519 And it's just been great, honestly because like I don't 987 00:47:22,519 --> 00:47:25,719 have to worry about where on the actual connectors anymore 988 00:47:25,719 --> 00:47:28,079 because I never get pulled in or pushed out. I 989 00:47:28,119 --> 00:47:29,639 had this problem, like every single time I went in 990 00:47:29,679 --> 00:47:31,840 the airplane. I have those one of those really annoying 991 00:47:32,599 --> 00:47:36,079 mini phono three point five millimeter like splitters in the 992 00:47:36,119 --> 00:47:39,280 plane just from my headphones, and I like always break 993 00:47:39,320 --> 00:47:41,719 them on the plane, like I will bash into them whatever, 994 00:47:41,760 --> 00:47:45,559 because you know the air airplanes famous for giving you 995 00:47:45,679 --> 00:47:47,639 lots of room to move around in I. 996 00:47:48,360 --> 00:47:50,880 Speaker 2: Have ruined them, and like this, I just it would 997 00:47:50,880 --> 00:47:51,480 just come off. 998 00:47:51,519 --> 00:47:57,840 Speaker 4: Honestly, it reminds me of the old Mac connectors. It's 999 00:47:57,880 --> 00:48:01,280 not USBC, it's not lightning, but it was like if 1000 00:48:01,280 --> 00:48:03,519 you toggled it just a little bit, it would come 1001 00:48:03,559 --> 00:48:06,320 off as opposed to like breaking off in the connector. 1002 00:48:06,440 --> 00:48:07,679 Speaker 3: Right, do you remember that? 1003 00:48:08,360 --> 00:48:09,960 Speaker 2: So I don't own I'm back, Okay. 1004 00:48:10,880 --> 00:48:14,639 Speaker 1: I was somewhat jealous of people with the ac adapter 1005 00:48:14,719 --> 00:48:16,400 connector that I think it was mag safe or something 1006 00:48:16,440 --> 00:48:20,599 like that, yeah to connect Yeah, And I don't know 1007 00:48:20,639 --> 00:48:22,280 why Apple got rid of it and then like brought 1008 00:48:22,280 --> 00:48:23,320 it back like that. 1009 00:48:23,360 --> 00:48:24,639 Speaker 2: It seemed like they were onto something. 1010 00:48:24,880 --> 00:48:28,320 Speaker 1: I do get that it falls off, h but for me, 1011 00:48:28,440 --> 00:48:31,760 like I'm not using it for like walking around with 1012 00:48:32,119 --> 00:48:35,440 so it's like my USB key. I'm on, I'm traveling somewhere. 1013 00:48:35,480 --> 00:48:37,239 It's really annoying to stick it into my laptop and 1014 00:48:37,239 --> 00:48:39,559 pull it back out again. This is like really easy swap. 1015 00:48:39,639 --> 00:48:41,000 I don't know what it was that I just like 1016 00:48:41,039 --> 00:48:42,840 clicked for me, like I could actually do this. I 1017 00:48:43,039 --> 00:48:45,119 guess I never wanted to do it with like USBA 1018 00:48:45,440 --> 00:48:47,679 because it just seemed like such a legacy technology all 1019 00:48:47,719 --> 00:48:51,159 the time. But now now that everything's USBC, like this 1020 00:48:51,239 --> 00:48:53,920 is this has been a Yeah, absolutely fantastic. 1021 00:48:54,159 --> 00:48:55,239 Speaker 3: I'm gonna have to check it out. 1022 00:48:56,480 --> 00:48:59,199 Speaker 1: Yeah, be prepared to uh. I think this is like 1023 00:48:59,280 --> 00:49:02,880 by Han send Da or something like that. I don't know, 1024 00:49:02,880 --> 00:49:06,159 some random brand on Amazon. I'm sure it's ripped off 1025 00:49:06,199 --> 00:49:08,280 from like another company that makes really good ones. 1026 00:49:08,320 --> 00:49:09,840 Speaker 2: These were actually not cheap. 1027 00:49:09,599 --> 00:49:12,320 Speaker 3: Though, So it's the same one, just a different logo. 1028 00:49:12,800 --> 00:49:15,320 Speaker 2: Oh yeah, that's the thing. There's no there's no even 1029 00:49:15,320 --> 00:49:16,000 I don't know if there's a. 1030 00:49:16,000 --> 00:49:19,519 Speaker 1: Logo on it actually, so like that's the most suspicious part. 1031 00:49:20,119 --> 00:49:21,800 Speaker 2: So that's gonna be my pick. 1032 00:49:22,800 --> 00:49:24,360 Speaker 1: You can't just buy one, though, you have to like 1033 00:49:24,400 --> 00:49:26,760 go fall in because like all your power adapters and 1034 00:49:26,800 --> 00:49:28,599 everything have to be connected otherwise. 1035 00:49:28,239 --> 00:49:29,719 Speaker 2: You're like, well, I have to pull out the plug 1036 00:49:30,000 --> 00:49:30,519 in order. 1037 00:49:30,440 --> 00:49:33,840 Speaker 1: To switch it. So be prepared if that's going to 1038 00:49:33,920 --> 00:49:37,639 be your future. All right, Okay, Jenna, what did you 1039 00:49:37,639 --> 00:49:38,239 bring for us? 1040 00:49:38,639 --> 00:49:42,360 Speaker 4: My pick is I'll give you a little backstory first. 1041 00:49:43,159 --> 00:49:48,320 I have never really understood the allure of mechanical keyboards. 1042 00:49:49,880 --> 00:49:53,920 Back before microphones actually had like noise canceling on that. 1043 00:49:54,039 --> 00:49:55,000 Speaker 3: My mind's right here. 1044 00:49:55,039 --> 00:49:59,360 Speaker 4: So back before microphones had noise canceling on them, they 1045 00:49:59,400 --> 00:50:02,960 were very and clickie and clackety, and I just. 1046 00:50:02,960 --> 00:50:04,039 Speaker 3: Didn't understand them. 1047 00:50:04,199 --> 00:50:07,039 Speaker 4: And anyways, a couple of years ago, I bought one 1048 00:50:07,119 --> 00:50:10,239 and fell in love, and I will show it to you. 1049 00:50:10,280 --> 00:50:10,880 Speaker 3: It's right here. 1050 00:50:11,440 --> 00:50:14,039 Speaker 4: This is not my This one specifically is not my pick. 1051 00:50:14,119 --> 00:50:18,280 But a couple of weeks ago, I bought another one 1052 00:50:18,920 --> 00:50:22,239 because well, first of all, I I occasionally have wrist pain, 1053 00:50:22,320 --> 00:50:25,840 so I wanted a different layout, and I also wanted 1054 00:50:25,840 --> 00:50:29,760 a project, and so I bought one that you actually 1055 00:50:29,760 --> 00:50:32,400 had to put together. And so I'm I'm only part 1056 00:50:32,440 --> 00:50:34,960 way through this. Like I said, I wanted, I needed 1057 00:50:35,000 --> 00:50:36,920 a project, like I spent enough time in front of 1058 00:50:36,920 --> 00:50:39,239 my screen, like I could actually build something with my 1059 00:50:39,400 --> 00:50:42,639 physical like physical hands. But it's it's kind of nice 1060 00:50:42,679 --> 00:50:47,760 to just have something something else that is not writing 1061 00:50:47,800 --> 00:50:51,039 code or being on a computer, but will still support 1062 00:50:51,079 --> 00:50:54,599 my computer years. So I got, I got a different layout, 1063 00:50:54,679 --> 00:50:57,119 I got you know, the key caps. I got different 1064 00:50:57,239 --> 00:50:59,400 switches that I think they're a little bit different than 1065 00:50:59,440 --> 00:51:02,039 the ones I have here. So I'm excited to kind 1066 00:51:02,039 --> 00:51:04,039 of try it out and figure out if I like 1067 00:51:04,079 --> 00:51:06,079 that this new one as much as I like this one. 1068 00:51:06,599 --> 00:51:08,079 Speaker 1: I thought for sure you're going to say that you 1069 00:51:08,159 --> 00:51:10,760 bought it and you still don't understand why people like 1070 00:51:10,800 --> 00:51:11,840 mechanical keyboards. 1071 00:51:11,960 --> 00:51:17,039 Speaker 4: Well, no, I mean, I'm definitely not like a fanatic 1072 00:51:17,119 --> 00:51:19,159 about it, like people are obsessed. 1073 00:51:19,280 --> 00:51:21,039 Speaker 1: Like, I'll be careful if you say that, We're going 1074 00:51:21,079 --> 00:51:22,840 to lose some viewers if you if you. 1075 00:51:23,320 --> 00:51:24,599 Speaker 3: No judgment, No judgment. 1076 00:51:25,199 --> 00:51:28,159 Speaker 4: When I was at Amazon, I wrote a blog post 1077 00:51:29,000 --> 00:51:32,400 about about mechanical keyboards. I just kind of wanted to 1078 00:51:32,480 --> 00:51:35,559 understand like who uses them, like who likes them? And 1079 00:51:35,599 --> 00:51:38,199 which one do you have? And got a lot of 1080 00:51:38,239 --> 00:51:40,559 a lot of a lot of opinions on there. And 1081 00:51:40,760 --> 00:51:44,360 one of my friends commented, coworkers commented, and he's like, 1082 00:51:44,400 --> 00:51:46,639 as soon as you start like building your own, then 1083 00:51:46,840 --> 00:51:48,440 like you're you've gone too far. 1084 00:51:49,079 --> 00:51:50,840 Speaker 3: So I've gone too far. 1085 00:51:51,960 --> 00:51:55,079 Speaker 1: I mean, I get I get the organomics for if 1086 00:51:55,119 --> 00:51:57,039 you have some sort of physical pain, like you know 1087 00:51:57,079 --> 00:51:59,159 that there's something wrong with what you're doing or if 1088 00:51:59,199 --> 00:52:00,800 you I mean, it took me a lot of years 1089 00:52:00,800 --> 00:52:02,880 to realize, oh wait, my whole job I revolves are 1090 00:52:02,920 --> 00:52:05,679 on my keyboard, I probably should have one that it's 1091 00:52:05,800 --> 00:52:08,159 me as best as I can, and that includes for 1092 00:52:08,239 --> 00:52:09,360 me it was a keyboard layout. 1093 00:52:09,599 --> 00:52:11,920 Speaker 2: I get totally the physical. 1094 00:52:11,519 --> 00:52:13,360 Speaker 1: Result, like actually having to push the keys, and whether 1095 00:52:13,440 --> 00:52:16,559 or not that's problematic. I have a thing for sound, though, 1096 00:52:17,039 --> 00:52:19,079 and so I did some research on like getting the 1097 00:52:19,159 --> 00:52:23,599 quietest keyboard possible, and someone a bunch of people said, oh, 1098 00:52:23,679 --> 00:52:27,599 they make really quiet mechanical keyboards, and I'm like, okay, sure. 1099 00:52:27,920 --> 00:52:30,000 So I probably spent like a whole bunch of hours 1100 00:52:30,000 --> 00:52:32,960 going around to different shops pushing the keys on different 1101 00:52:32,960 --> 00:52:36,639 mechanical keyboards in multiple countries, and like going online and 1102 00:52:36,679 --> 00:52:40,400 like listening to audio clips of the mechanical keyboards, and 1103 00:52:40,760 --> 00:52:43,559 after that, all I conclude is those people have no 1104 00:52:43,599 --> 00:52:44,599 idea what they're talking about. 1105 00:52:44,599 --> 00:52:47,800 Speaker 2: Mechanical keyboards are not quiet in any way. They're like 1106 00:52:48,679 --> 00:52:49,760 the quieter. 1107 00:52:49,480 --> 00:52:53,119 Speaker 1: Switches, like they're not They're not quiet. And I actually 1108 00:52:53,239 --> 00:52:55,599 my pick actually in a previous episode was my keyboard, 1109 00:52:55,639 --> 00:52:59,559 which is dedicated to be a silent keyboard, so people 1110 00:52:59,639 --> 00:53:01,960 can hit all they want for that. But I still 1111 00:53:02,000 --> 00:53:04,679 will not get mechanical keyboards unless you have a physical ailment, 1112 00:53:06,599 --> 00:53:08,119 because then you know, I get it, I tually do. 1113 00:53:10,039 --> 00:53:14,320 Speaker 4: Yeah, I also went down both times. I went down 1114 00:53:14,320 --> 00:53:16,119 a rabbit hole trying to figure out, like which one 1115 00:53:16,119 --> 00:53:16,599 should I get. 1116 00:53:16,719 --> 00:53:18,239 Speaker 3: I didn't go quite that far. I didn't. 1117 00:53:18,280 --> 00:53:21,280 Speaker 4: I didn't actually go to physical stores like we just don't. 1118 00:53:21,719 --> 00:53:24,559 There's not many here that sell what I was looking 1119 00:53:24,599 --> 00:53:27,679 for here. But you're right, there's still not quiet even 1120 00:53:27,719 --> 00:53:29,719 though like I got. I think I got like one 1121 00:53:29,760 --> 00:53:33,400 of the quieter sets of switches, and that the smoother 1122 00:53:33,519 --> 00:53:36,480 ones that don't make the as much of the clickity clackity. 1123 00:53:38,239 --> 00:53:39,800 It's there's still some sound there. 1124 00:53:39,880 --> 00:53:41,840 Speaker 1: So I have to I have to ask you, So, 1125 00:53:42,039 --> 00:53:45,159 what's the keyboard brander model that, yeah, you purchased. 1126 00:53:45,559 --> 00:53:48,079 Speaker 4: So, so the one, the one that I showed you 1127 00:53:48,239 --> 00:53:51,679 is a key cron Q one pro wireless. It's just 1128 00:53:51,800 --> 00:53:55,480 is I think it's like the seventy five percent. I'm 1129 00:53:55,480 --> 00:53:56,519 holding it up for the people. 1130 00:53:56,519 --> 00:53:57,079 Speaker 2: They look small. 1131 00:53:57,199 --> 00:54:00,000 Speaker 4: Yeah, it doesn't have like the number pad on the side, 1132 00:54:00,559 --> 00:54:03,039 so it's the seventy five percent layout. I think the 1133 00:54:03,079 --> 00:54:05,199 new one that I got is an Alice layout, so 1134 00:54:05,239 --> 00:54:08,760 it's a little bit more rounded, so that where your 1135 00:54:08,840 --> 00:54:12,360 your hands are I'm rotating my hands where your hands 1136 00:54:12,360 --> 00:54:14,320 are a little bit in a more natural layout as 1137 00:54:14,320 --> 00:54:17,079 opposed to straight up and down like you would on 1138 00:54:17,239 --> 00:54:18,679 kind of a regular keyboard. 1139 00:54:18,760 --> 00:54:20,199 Speaker 3: And it's a little bit bigger too. 1140 00:54:20,559 --> 00:54:22,440 Speaker 4: One thing that like I've noticed with this one here 1141 00:54:23,199 --> 00:54:25,880 is it is a it's narrower. It's seventy five percent, 1142 00:54:25,920 --> 00:54:28,239 so it's narrower, and so like my hands are closer together. 1143 00:54:28,400 --> 00:54:30,880 And the one other piece that I like about this 1144 00:54:30,960 --> 00:54:35,920 brand specifically or this this these is that it's pretty hefty, 1145 00:54:36,039 --> 00:54:41,000 like this is several pounds and so it's it's solid. 1146 00:54:41,119 --> 00:54:42,760 I don't know, I just I like the feel of it. 1147 00:54:42,760 --> 00:54:44,920 Speaker 2: It feels like it's real, a post fake. 1148 00:54:45,000 --> 00:54:47,079 Speaker 3: Yeah, yeah, it's not going to go anywhere if I 1149 00:54:47,119 --> 00:54:48,760 get like carried away. 1150 00:54:50,280 --> 00:54:52,639 Speaker 1: I think we have probably like one guest per year 1151 00:54:52,719 --> 00:54:55,159 on the show who calls out their mechanical keyboard. 1152 00:54:55,239 --> 00:54:56,079 Speaker 4: Is there? Is there? Pick? 1153 00:54:56,199 --> 00:54:58,480 Speaker 2: So you're you're in good company, Okay, good? 1154 00:54:58,840 --> 00:55:02,199 Speaker 1: Yeah, So that I think we'll call it the end 1155 00:55:02,199 --> 00:55:04,920 of the episode there. Thank you so much Jenna for 1156 00:55:05,079 --> 00:55:10,360 coming on and being our our target practice for today's episode. 1157 00:55:10,679 --> 00:55:12,599 Speaker 3: Thank you so much for having me. It's been it's 1158 00:55:12,599 --> 00:55:13,480 been a good conversation. 1159 00:55:14,239 --> 00:55:15,960 Speaker 1: It has been I've enjoyed it, and I just want 1160 00:55:16,000 --> 00:55:18,960 to thank personally attribute one last time for sponsoring today's episode. 1161 00:55:19,039 --> 00:55:20,239 Speaker 2: And I hope to see all 1162 00:55:20,280 --> 00:55:24,320 Speaker 1: Of our viewers and listeners next week