1
00:00:04,409 --> 00:00:08,050
OpenAI AI model solves 80-year math conjecture.

2
00:00:08,409 --> 00:00:12,149
Gower's signing off on the proof changes what counts as frontier research.

3
00:00:12,409 --> 00:00:16,450
Labs now face pressure to integrate these models into every conjecture pipeline

4
00:00:16,569 --> 00:00:18,840
or watch their publication lead evaporate.

5
00:00:19,129 --> 00:00:20,149
Within 18 months,

6
00:00:20,209 --> 00:00:23,149
the first AI co-authored paper in a top math journal

7
00:00:23,290 --> 00:00:25,430
should outside its human-only peers.

8
00:00:25,649 --> 00:00:27,479
That shift pulls grant money away from

9
00:00:27,479 --> 00:00:30,370
traditional theory groups toward compute budgets instead.

10
00:00:30,770 --> 00:00:33,409
Smaller departments lose their industry edge first.

11
00:00:33,830 --> 00:00:37,599
Anthropic commits $45 billion to SpaceX for compute.

12
00:00:37,950 --> 00:00:41,349
Locking in that monthly outlay through 2029 leaves little room

13
00:00:41,349 --> 00:00:44,430
for pivots if inference costs drop faster than projected.

14
00:00:44,860 --> 00:00:47,630
The deal forces OpenAI to either match with its own

15
00:00:47,630 --> 00:00:51,540
satellite cluster or accept higher latency on frontier runs.

16
00:00:51,830 --> 00:00:54,110
Smaller players without such contracts face

17
00:00:54,110 --> 00:00:56,849
immediate capacity rationing starting next quarter.

18
00:00:57,150 --> 00:01:00,310
This arrangement signals that power availability on Earth has

19
00:01:00,310 --> 00:01:02,950
become the binding constraint for every major lab.

20
00:01:03,229 --> 00:01:04,309
Watch as your margins.

21
00:01:04,519 --> 00:01:06,879
Compress one orbital capacity scales.

22
00:01:07,279 --> 00:01:10,459
Anthropic's Jack Clark forecasts Nobel AI win.

23
00:01:10,709 --> 00:01:13,120
Clark's timeline for recursive self-improvement

24
00:01:13,120 --> 00:01:15,680
by late 2028 undercuts the slower

25
00:01:15,680 --> 00:01:18,980
roadmap still published by safety teams at the same labs.

26
00:01:19,239 --> 00:01:21,839
Expect grant reviewers to start requiring AI

27
00:01:21,839 --> 00:01:23,940
co-authors on proposals within the year.

28
00:01:24,269 --> 00:01:27,419
That change sidelines purely human-led labs first.

29
00:01:27,650 --> 00:01:32,000
OpenAI safety staff must now either publish matching forecasts or

30
00:01:32,000 --> 00:01:34,660
explain why their models will lag clawed on autonomy.

31
00:01:35,330 --> 00:01:38,209
University departments ignoring the shift lose top graduate

32
00:01:38,209 --> 00:01:41,349
students to compute heavy groups by fall 2026.

33
00:01:41,730 --> 00:01:44,830
Meta trained AI on employee data before layoffs.

34
00:01:45,010 --> 00:01:47,209
The timing with those layoffs suggests the models

35
00:01:47,209 --> 00:01:50,160
were trained precisely to replace the tracked workflows.

36
00:01:50,489 --> 00:01:53,930
Google employees now have fresh reason to push for explicit data

37
00:01:53,930 --> 00:01:57,449
use audits before their own activity feeds the next Gemini iteration.

38
00:01:58,165 --> 00:01:59,375
Without those protections,

39
00:01:59,574 --> 00:02:02,855
top engineers start filtering their logs or leaving for smaller

40
00:02:02,855 --> 00:02:05,794
shops that cannot yet afford the same surveillance stack.

41
00:02:06,044 --> 00:02:08,975
Expect similar leaks from Amazon within 6 months

42
00:02:08,975 --> 00:02:11,475
as infrastructure bills mount across the sector.

43
00:02:11,854 --> 00:02:14,395
Trump scraps AI safety executive order.

44
00:02:14,695 --> 00:02:17,535
Direct lobbying from the top 3 AI CEOs

45
00:02:17,535 --> 00:02:20,035
killed the Frontier Review before ink hit paper.

46
00:02:20,330 --> 00:02:23,050
The move speeds US model releases by months,

47
00:02:23,279 --> 00:02:27,800
but it invites allies to impose their own pre-market rules on American exports.

48
00:02:28,130 --> 00:02:31,830
This forces OpenAI to build an internal certification process

49
00:02:31,970 --> 00:02:35,360
that mirrors the canceled order within 18 months or see its

50
00:02:35,360 --> 00:02:37,570
enterprise pipeline shift toward regulated

51
00:02:37,570 --> 00:02:39,839
domestic alternatives in key markets.

52
00:02:40,210 --> 00:02:44,550
Google and Meta must publish voluntary audit reports ahead of any foreign mandate.

53
00:02:44,929 --> 00:02:45,289
GitHub.

54
00:02:45,429 --> 00:02:47,929
By poisoned VS code extension hack.

55
00:02:48,169 --> 00:02:50,830
18 minutes was enough for the poisoned extension to

56
00:02:50,830 --> 00:02:54,139
drain thousands of repos and targeted AI credentials.

57
00:02:54,429 --> 00:02:57,029
Those clawed configs now let attackers test

58
00:02:57,029 --> 00:02:59,779
prompt injections against live enterprise accounts.

59
00:03:00,070 --> 00:03:02,949
This forces GitHub to require cryptographic signing

60
00:03:02,949 --> 00:03:05,369
for all marketplace extensions by year end

61
00:03:05,509 --> 00:03:07,789
or lose the trust of security conscious AI

62
00:03:07,789 --> 00:03:10,410
labs that maintain their own internal stores instead.

63
00:03:10,889 --> 00:03:14,369
Developers everywhere will start pinning extension versions and blocking

64
00:03:14,369 --> 00:03:17,479
auto updates after seeing how fast the window opened.

65
00:03:17,960 --> 00:03:19,970
Supply chain worm strikes OpenAI,

66
00:03:20,279 --> 00:03:20,679
Mistral,

67
00:03:20,729 --> 00:03:21,389
and more.

68
00:03:21,690 --> 00:03:26,550
The NPM packages sat in 170 places before anyone spotted the payload.

69
00:03:26,889 --> 00:03:30,490
OpenAI and Mistral employee machines now carry persistent

70
00:03:30,490 --> 00:03:33,169
access that can siphon model configs over time.

71
00:03:33,509 --> 00:03:35,669
This forces both labs to rebuild their

72
00:03:35,669 --> 00:03:38,190
entire dependency trees from clean sources within

73
00:03:38,190 --> 00:03:43,259
30 days or risk gradual weight exfiltration that no perimeter tool will catch.

74
00:03:43,669 --> 00:03:46,869
Teams at the European Commission face the same rebuild cycle,

75
00:03:47,029 --> 00:03:50,550
which will delay their AI procurement pilots by at least 1 quarter.

76
00:03:51,190 --> 00:03:54,529
Google IO unveils broad agentic AI ecosystem.

77
00:03:54,979 --> 00:03:57,660
The always on agents turn every workspace session

78
00:03:57,660 --> 00:04:00,250
into a running inference job that never sleeps.

79
00:04:00,619 --> 00:04:03,580
Salesforce loses its mid-market wedge by Q3

80
00:04:03,740 --> 00:04:05,979
once Google agents start completing the same

81
00:04:05,979 --> 00:04:08,850
CRM updates inside existing email threads.

82
00:04:09,100 --> 00:04:10,259
Enterprise IT team.

83
00:04:10,464 --> 00:04:14,085
will now demand usage-based credits instead of seat licenses,

84
00:04:14,125 --> 00:04:16,945
squeezing margins on any standalone agent platform

85
00:04:16,945 --> 00:04:19,415
that can't embed inside Gmail and Docs.

86
00:04:19,744 --> 00:04:22,904
Google's move also locks in device data flows that Apple

87
00:04:22,904 --> 00:04:25,704
cannot match without its own OS level agent layer.

88
00:04:27,070 --> 00:04:30,279
Chat GPT gains PowerPoint generation capability.

89
00:04:30,510 --> 00:04:34,899
Agencies lose their billable hours the moment brand voice lives inside the model.

90
00:04:35,230 --> 00:04:38,450
Chat GPT's deck builder pulls from conversation history.

91
00:04:38,709 --> 00:04:42,829
So one prompt now spits out a 20 slide deck with charts and speaker notes.

92
00:04:43,109 --> 00:04:46,390
That changes the economics for anyone selling templates or training.

93
00:04:46,790 --> 00:04:50,190
Mid-market slide software faces churn as teams test whether

94
00:04:50,190 --> 00:04:53,399
the output meets compliance without a human review layer.

95
00:04:53,609 --> 00:04:57,950
Expect template marketplaces to pivot toward verification services by year-end

96
00:04:58,109 --> 00:05:00,029
or watch usage metrics flatline.

97
00:05:00,760 --> 00:05:03,739
AI sparks surge in lawsuits and creative output.

98
00:05:04,000 --> 00:05:07,399
The litigation spike distracts from how fast publishers are locking

99
00:05:07,399 --> 00:05:10,220
in data licensing deals to feed the same models.

100
00:05:10,480 --> 00:05:15,239
AI generated papers now outnumber human submissions in some preprint servers,

101
00:05:15,359 --> 00:05:18,739
yet the lawsuits target the training sets rather than the outputs.

102
00:05:18,940 --> 00:05:19,679
That mismatch.

103
00:05:19,829 --> 00:05:24,130
Academic platforms will add watermarking requirements before journals catch up.

104
00:05:24,429 --> 00:05:27,390
Smaller research tools without provenance tracking lose

105
00:05:27,390 --> 00:05:30,269
their edge once institutions demand audit trails.

106
00:05:30,540 --> 00:05:33,230
API margins compress once compliance checks

107
00:05:33,230 --> 00:05:35,670
become mandatory for every generated citation.

108
00:05:36,070 --> 00:05:39,070
AI robots outpace humans in tire changes.

109
00:05:39,359 --> 00:05:42,519
Repair shops choose between retrofitting for robots or ceding

110
00:05:42,519 --> 00:05:45,350
tire jobs to fleets already testing the systems.

111
00:05:45,589 --> 00:05:49,149
Robots complete a swap in under 4 minutes with zero torque errors,

112
00:05:49,359 --> 00:05:51,459
which undercuts the labor model at independent.

113
00:05:51,640 --> 00:05:52,410
Garages.

114
00:05:52,630 --> 00:05:55,429
Insurance carriers will start offering lower premiums

115
00:05:55,429 --> 00:05:58,109
only to locations that log robotic assistance,

116
00:05:58,309 --> 00:06:00,910
pushing adoption faster than any hardware sale.

117
00:06:01,230 --> 00:06:03,980
Traditional toolmakers must bundle vision systems

118
00:06:04,109 --> 00:06:07,209
or see their sockets and lifts gather dust by next summer.

119
00:06:07,829 --> 00:06:11,109
US-China AI safety talks advance amid competition.

120
00:06:11,470 --> 00:06:15,010
Both sides keep the compute export rules exactly where they are,

121
00:06:15,149 --> 00:06:17,709
while the safety agenda moves forward on paper.

122
00:06:18,029 --> 00:06:22,190
Frontier labs on each coast will still race to the next token threshold

123
00:06:22,190 --> 00:06:26,779
because the talks focus on evaluation standards rather than shared compute caps.

124
00:06:27,109 --> 00:06:29,390
Model developers gain another quarter before

125
00:06:29,390 --> 00:06:31,850
any hardware restriction actually binds.

126
00:06:32,070 --> 00:06:35,269
That leaves smaller nations outside the conversation forced

127
00:06:35,269 --> 00:06:37,269
to pick one supply chain or the other.

128
00:06:37,549 --> 00:06:40,140
The real constraint emerges when audit requirements

129
00:06:40,140 --> 00:06:42,630
hit the inference providers serving both markets.