WEBVTT

1
00:00:00.040 --> 00:00:04.480
<v Speaker 1>Ai Daily Briefing. I'm ad Sharp, thanks for joining me today.

2
00:00:04.759 --> 00:00:09.480
<v Speaker 1>Open AI's three tier strategy. Open ai just restructured how

3
00:00:09.480 --> 00:00:13.160
<v Speaker 1>it sells intelligence. GPT Dash five point six is out today,

4
00:00:13.359 --> 00:00:15.519
<v Speaker 1>and it doesn't come as a single model. It comes

5
00:00:15.640 --> 00:00:19.519
<v Speaker 1>is three Sol, Terror and Luna, each targeting a different

6
00:00:19.559 --> 00:00:23.800
<v Speaker 1>tier of cost, speed, and capability. That's a meaningful strategic shift,

7
00:00:24.039 --> 00:00:26.480
<v Speaker 1>and it's the clearest signal yet that open ai is

8
00:00:26.519 --> 00:00:30.440
<v Speaker 1>moving from model releases toward market segmentation. Here's what that

9
00:00:30.480 --> 00:00:35.079
<v Speaker 1>means in practice. Soul sits at the capable end, Terror

10
00:00:35.079 --> 00:00:37.679
<v Speaker 1>in the middle, and Luna at the lighter, faster end

11
00:00:37.719 --> 00:00:41.359
<v Speaker 1>of the spectrum. The tier naming is also new generation.

12
00:00:41.520 --> 00:00:45.880
<v Speaker 1>Numbers and capability names are now separated, so GPT five

13
00:00:45.920 --> 00:00:49.560
<v Speaker 1>point six is the version, and Soul, Terror Luna describe

14
00:00:49.600 --> 00:00:53.079
<v Speaker 1>where you land on the capability curve. GPT five point

15
00:00:53.079 --> 00:00:57.359
<v Speaker 1>four for reference retires July twenty third. The important distinction

16
00:00:57.679 --> 00:00:59.799
<v Speaker 1>is that this isn't just a pricing move, It's a

17
00:01:00.000 --> 00:01:03.679
<v Speaker 1>distribution bet. Open ai is wagering the cost of wear

18
00:01:03.719 --> 00:01:07.439
<v Speaker 1>adoption across enterprise and consumer segments will drive broader deployment

19
00:01:07.640 --> 00:01:11.200
<v Speaker 1>than a single flagship model ever could. Alongside the model launch.

20
00:01:11.439 --> 00:01:15.680
<v Speaker 1>Open Ai released the chat GPT work Agent today across web, mobile,

21
00:01:15.680 --> 00:01:20.439
<v Speaker 1>and desktop. It integrates Codex powered development capabilities directly into

22
00:01:20.480 --> 00:01:24.799
<v Speaker 1>the interface. The consolidation play here is clear, fewer separate

23
00:01:24.840 --> 00:01:29.319
<v Speaker 1>tools one surface. Whether existing plus and pro users migrate

24
00:01:29.359 --> 00:01:32.120
<v Speaker 1>to the work tier smoothly is still an open question.

25
00:01:32.640 --> 00:01:36.079
<v Speaker 1>The Codex transition path isn't fully settled, and adding new

26
00:01:36.120 --> 00:01:40.159
<v Speaker 1>capabilities to a product people already use in established ways

27
00:01:40.159 --> 00:01:43.560
<v Speaker 1>creates friction that isn't always visible at launch. Now to

28
00:01:43.599 --> 00:01:47.239
<v Speaker 1>the defense side, where the situation is moving faster than expected.

29
00:01:47.760 --> 00:01:51.719
<v Speaker 1>An AFARL memo dated July ninth accelerates the removal of

30
00:01:51.760 --> 00:01:55.599
<v Speaker 1>all Anthropic products from Air Force contractor systems to September first.

31
00:01:56.200 --> 00:01:59.040
<v Speaker 1>That's nearly a month ahead of the broader DoD deadline

32
00:01:59.079 --> 00:02:02.599
<v Speaker 1>of September twenty ninth. The signal here is that implementation

33
00:02:02.799 --> 00:02:06.480
<v Speaker 1>isn't waiting for legal resolution. Contractors are being pushed to

34
00:02:06.519 --> 00:02:09.039
<v Speaker 1>act before the courts of funished sorting out whether the

35
00:02:09.120 --> 00:02:13.680
<v Speaker 1>underlying ban holds. Anthropics sue to overturn the original Pentagon

36
00:02:13.719 --> 00:02:17.759
<v Speaker 1>decision and executive communications related to that dispute have recently

37
00:02:17.800 --> 00:02:22.879
<v Speaker 1>been released, which complicates the DoD's legal position. If Anthropic prevails,

38
00:02:23.240 --> 00:02:26.360
<v Speaker 1>agencies could be required to reverse the purge. That's a

39
00:02:26.400 --> 00:02:29.639
<v Speaker 1>real compliance risk contractors are currently absorbing with no clear

40
00:02:29.680 --> 00:02:34.960
<v Speaker 1>outcome guaranteed. Here's the thing. The original conflict is worth understanding. Precisely,

41
00:02:35.599 --> 00:02:39.360
<v Speaker 1>the Pentagon cited unacceptable supply chain risk language that was

42
00:02:39.439 --> 00:02:42.599
<v Speaker 1>later softened in an April memo. The core issue is

43
00:02:42.639 --> 00:02:46.639
<v Speaker 1>that anthropics safety guard rails limit certain use cases, including

44
00:02:46.680 --> 00:02:51.039
<v Speaker 1>surveillance and autonomous weapons applications. The DoD wants tools without

45
00:02:51.039 --> 00:02:55.439
<v Speaker 1>those constraints. Anthropis position is that those constraints exist for reasons.

46
00:02:56.000 --> 00:02:59.159
<v Speaker 1>Neither side has moved significantly, and the legal in compliance

47
00:02:59.199 --> 00:03:04.560
<v Speaker 1>cascades are still grinding forward. Separately, thedld's attempt to clarify

48
00:03:04.639 --> 00:03:08.599
<v Speaker 1>rules around Chinese military company contractors has produced more confusion,

49
00:03:08.800 --> 00:03:12.840
<v Speaker 1>not less. A class deviation on definitions leaves a critical

50
00:03:12.919 --> 00:03:16.879
<v Speaker 1>term undefined what counts as reasonable inquiry when vetting a

51
00:03:16.919 --> 00:03:21.680
<v Speaker 1>suppli's China connections. Contractors are now interpreting that standard themselves,

52
00:03:22.199 --> 00:03:25.039
<v Speaker 1>and the consequence of getting it wrong is false claims

53
00:03:25.039 --> 00:03:28.759
<v Speaker 1>act liability. That's an aggressive exposure for a rule that

54
00:03:28.800 --> 00:03:31.400
<v Speaker 1>still hasn't gone through the full notice and comment process.

55
00:03:31.879 --> 00:03:35.520
<v Speaker 1>On the industry side, healthcare AI startup EPEN evidence is

56
00:03:35.520 --> 00:03:40.120
<v Speaker 1>showing what domain specific design actually delivers. Its retrieval augmented

57
00:03:40.159 --> 00:03:44.719
<v Speaker 1>generation approach outperforms general large language models on medical accuracy

58
00:03:44.759 --> 00:03:49.439
<v Speaker 1>and evidence traceability. The clinician facing result is fewer hallucinations

59
00:03:49.639 --> 00:03:53.520
<v Speaker 1>and clearer sourcing. That matters because in high stakes environments,

60
00:03:53.759 --> 00:03:59.199
<v Speaker 1>conversational fluency isn't the goal. Verifiable traceable answers are general

61
00:03:59.199 --> 00:04:03.759
<v Speaker 1>models weren't built for that. Specialized ones increasingly are three

62
00:04:03.759 --> 00:04:07.240
<v Speaker 1>things to track from here. First, whether open ai'ed tiered

63
00:04:07.280 --> 00:04:10.319
<v Speaker 1>model structure drives the adoption breadth its designed for, or

64
00:04:10.360 --> 00:04:14.319
<v Speaker 1>whether it creates confusion rather than clarity. Second, whether the

65
00:04:14.319 --> 00:04:18.279
<v Speaker 1>anthropic lawsuit produces any interim rulings before the September first

66
00:04:18.319 --> 00:04:23.160
<v Speaker 1>Air Force deadline lands. Third, whether the DoD's undefined reasonable

67
00:04:23.160 --> 00:04:27.759
<v Speaker 1>inquiry standard gets clarified before false claims at cases start materializing.

68
00:04:28.279 --> 00:04:31.000
<v Speaker 1>The through line across today's stories is that AI diploma

69
00:04:31.199 --> 00:04:35.759
<v Speaker 1>is fragmenting by tier, bi sector, by legal jurisdiction. The

70
00:04:35.800 --> 00:04:38.439
<v Speaker 1>decisions made in the next sixty days across all three

71
00:04:38.439 --> 00:04:42.160
<v Speaker 1>of these tracks will shape how that fragmentation settles. Thanks

72
00:04:42.160 --> 00:04:45.720
<v Speaker 1>for listening. This podcast was built using AI technology, A

73
00:04:45.839 --> 00:04:46.680
<v Speaker 1>yes we production