1
00:00:04,320 --> 00:00:07,599
OpenAI rolls out trusted contact and chat GPT.

2
00:00:08,069 --> 00:00:12,819
Every intervention by a trusted contact feeds OpenAI a contextual refusal label

3
00:00:13,000 --> 00:00:15,539
that aborators cannot synthesize at scale.

4
00:00:15,760 --> 00:00:18,879
The feature turns oversight into preference data collection

5
00:00:19,079 --> 00:00:22,309
that competitors gather only through expensive contractors.

6
00:00:22,760 --> 00:00:25,860
Those signals compound across millions of family accounts

7
00:00:25,989 --> 00:00:28,360
and sharpen the model's ability to separate real

8
00:00:28,360 --> 00:00:31,799
distress from hypothetical queries without over refusing.

9
00:00:32,200 --> 00:00:33,139
Within 12 months,

10
00:00:33,560 --> 00:00:36,459
this pressures Anthropic to launch an equivalent feature

11
00:00:36,569 --> 00:00:39,560
or accept slower iteration on consumer workloads.

12
00:00:41,099 --> 00:00:43,889
EU agrees to dilute and delay AI rules.

13
00:00:44,259 --> 00:00:47,259
Countries just kicked enforcement to late 2027.

14
00:00:47,529 --> 00:00:50,220
Yet the real outcome is that European startups must

15
00:00:50,220 --> 00:00:54,209
now fine-tune American models rather than train sovereign ones.

16
00:00:54,540 --> 00:00:56,810
The extension gives Washington Labs time to

17
00:00:56,810 --> 00:01:00,090
lock enterprise workflows behind their APIs first.

18
00:01:00,540 --> 00:01:02,900
Compliance costs for high risk systems will

19
00:01:02,900 --> 00:01:05,260
hit after market share is already decided.

20
00:01:05,790 --> 00:01:09,650
This forces local players to embed US APIs deeply into their

21
00:01:09,650 --> 00:01:14,050
stacks or risk being outrun on capability before rules even apply.

22
00:01:15,000 --> 00:01:17,800
Google tests REMI autonomous agent in Gemini.

23
00:01:18,199 --> 00:01:20,839
REMI learns preferences by observing actual app

24
00:01:20,839 --> 00:01:24,019
interactions instead of waiting for explicit instructions.

25
00:01:24,360 --> 00:01:27,400
That observation data builds a behavior graph no

26
00:01:27,400 --> 00:01:29,379
prompt engineer can match after the fact.

27
00:01:29,959 --> 00:01:33,519
Once it starts handling cross-app tasks like Bookings or scheduling,

28
00:01:33,849 --> 00:01:37,949
Google captures every API call that third party services expose.

29
00:01:38,330 --> 00:01:41,370
This forces OpenAI and Anthropic to either replicate

30
00:01:41,370 --> 00:01:44,910
the same multi-app orchestration layer inside their agents,

31
00:01:45,089 --> 00:01:49,750
or watch daily active workflows migrate to Gemini by early 2026.

32
00:01:51,110 --> 00:01:54,129
Meta develops advanced genic AI Assistant.

33
00:01:54,349 --> 00:01:57,029
The shopping features inside Instagram turn every

34
00:01:57,029 --> 00:01:59,970
agent conversation into a direct checkout path

35
00:02:00,110 --> 00:02:03,089
that bypasses traditional web funnels entirely.

36
00:02:03,430 --> 00:02:06,910
Muse Spark reuses existing recommendation weights to drive

37
00:02:06,910 --> 00:02:09,970
those autonomous decisions without fresh training runs,

38
00:02:10,229 --> 00:02:11,889
keeping compute costs low.

39
00:02:12,110 --> 00:02:15,460
Expect this to accelerate Meta's ad attribution advantage

40
00:02:15,460 --> 00:02:17,990
as agent actions become the new click stream.

41
00:02:18,500 --> 00:02:22,000
This pushes Shopify to either integrate Meta's agent stack

42
00:02:22,179 --> 00:02:24,660
or watch its merchant tools lose relevance for

43
00:02:24,660 --> 00:02:26,699
younger buyers who never leave the app.

44
00:02:28,429 --> 00:02:31,660
Coder launches self-hosted model agnostic agents.

45
00:02:32,070 --> 00:02:35,710
Teams can now route developer tasks across any model while

46
00:02:35,710 --> 00:02:38,910
keeping every line of code inside their own VPCs.

47
00:02:39,270 --> 00:02:42,100
The beta ships with orchestration that swaps providers

48
00:02:42,100 --> 00:02:44,410
on the fly based on task type,

49
00:02:44,550 --> 00:02:47,779
and keeps latency under 2 seconds on local clusters.

50
00:02:48,190 --> 00:02:51,309
Mid-market software groups testing this setup report cutting

51
00:02:51,309 --> 00:02:54,839
inference spend by 33% without sacrificing accuracy.

52
00:02:55,360 --> 00:02:59,199
GitHub must expose similar self-hosted routing in co-pilot

53
00:02:59,320 --> 00:03:02,559
or lose enterprise deals to data residency demands.

54
00:03:04,860 --> 00:03:08,850
11 labs hits $500 million ARR milestone.

55
00:03:09,240 --> 00:03:13,729
Voice models just became the highest margin product line in generative AI.

56
00:03:14,070 --> 00:03:17,429
The milestone reveals that synthetic speech now handles more

57
00:03:17,429 --> 00:03:20,300
than half of all new customer service deployments.

58
00:03:20,710 --> 00:03:23,009
Latency improvements in the latest release

59
00:03:23,190 --> 00:03:26,190
make real-time conversation viable for the first time.

60
00:03:26,690 --> 00:03:30,300
Existing audio platforms must either embed similar generation

61
00:03:30,889 --> 00:03:33,559
or watch their per minute margins get undercut by

62
00:03:33,559 --> 00:03:36,919
50% as clients demand natural voices at scale.

63
00:03:37,289 --> 00:03:40,690
Procurement teams are already rewriting RFPs around voice

64
00:03:40,690 --> 00:03:42,889
quality scores instead of price per minute.

65
00:03:45,149 --> 00:03:48,100
OpenAI launches voice API for developers.

66
00:03:48,509 --> 00:03:51,270
Real-time voice just became another API call,

67
00:03:51,509 --> 00:03:54,309
yet the per minute pricing still assumes text-level

68
00:03:54,309 --> 00:03:57,449
inference costs that won't hold one usage scales.

69
00:03:57,750 --> 00:04:02,610
Early testers report subsecond latency even on complex multi-turn dialogues.

70
00:04:02,869 --> 00:04:05,520
The real Constraint surfaces when these calls run

71
00:04:05,520 --> 00:04:09,500
inside existing enterprise SLAs that penalize any jitter.

72
00:04:09,759 --> 00:04:12,000
Expect rate card revisions by summer.

73
00:04:12,399 --> 00:04:16,980
DeepGram and Assembly AI must either match the latency at comparable rates

74
00:04:17,000 --> 00:04:21,380
or accept they lose the developer mindshare they built on audio-only pipelines.

75
00:04:21,880 --> 00:04:26,079
SAP acquires prior labs for 1.18B AI lab.

76
00:04:26,519 --> 00:04:29,100
Frontier Model Development just moved inside

77
00:04:29,100 --> 00:04:31,769
the largest ERP vendor's budget cycle.

78
00:04:32,040 --> 00:04:36,190
The lab will prioritize fine-tuning on supply chain and finance workflows,

79
00:04:36,440 --> 00:04:39,450
where proprietary data already lives in their systems.

80
00:04:39,720 --> 00:04:42,940
Integration timelines point to general availability inside.

81
00:04:43,075 --> 00:04:45,255
S4 HANA by late next year.

82
00:04:45,595 --> 00:04:49,195
Competitors like Salesforce now face bundled AI that ships with

83
00:04:49,195 --> 00:04:52,174
the core license at no extra per user cost.

84
00:04:52,545 --> 00:04:56,614
This forces Salesforce to either slash Einstein fees or accelerate

85
00:04:56,614 --> 00:04:59,575
its own model partnerships before the next fiscal year.

86
00:05:00,890 --> 00:05:05,190
OpenAI and Oracle signed $100 billion supercluster deal.

87
00:05:05,489 --> 00:05:08,970
Oracle just secured routing control over OpenAI's largest

88
00:05:08,970 --> 00:05:11,359
inference pool for the next 18 months.

89
00:05:11,730 --> 00:05:13,529
Training jobs stay on Microsoft,

90
00:05:13,730 --> 00:05:16,029
but inference traffic splits permanently.

91
00:05:16,410 --> 00:05:20,049
That creates a direct margin squeeze on Azure's AI services

92
00:05:20,369 --> 00:05:23,950
because it can no longer bundle every token through its own fabric.

93
00:05:24,209 --> 00:05:27,410
The first sign appears in enterprise renewal negotiations,

94
00:05:27,609 --> 00:05:31,429
where customers start demanding multi-cloud credits to avoid lock-in.

95
00:05:31,730 --> 00:05:35,170
Nvidia watches closely since Oracle's cluster design

96
00:05:35,170 --> 00:05:38,109
favors AMD accelerators in the initial phase.

97
00:05:39,089 --> 00:05:42,380
DeepMind scales alpha evolve for protein folding.

98
00:05:42,940 --> 00:05:46,220
Alpha evolve's quantum integration cuts stimulation times

99
00:05:46,220 --> 00:05:49,420
for molecular interactions from weeks to hours.

100
00:05:49,869 --> 00:05:53,720
Drug developers gain the ability to test variant stability in-house

101
00:05:53,940 --> 00:05:56,600
rather than waiting on external supercomputers.

102
00:05:56,899 --> 00:06:02,359
That compresses early discovery from 18 months to under 9 for many protein targets.

103
00:06:02,630 --> 00:06:08,000
Alumina now has to embed comparable folding APIs into its sequencing platforms

104
00:06:08,179 --> 00:06:12,420
or risk losing workflow lock-in as labs migrate analysis upstream.

105
00:06:12,890 --> 00:06:15,609
Smaller biotechs without quantum access fall

106
00:06:15,609 --> 00:06:17,950
further behind on competitive timelines.

107
00:06:18,940 --> 00:06:22,190
Apple R&D spending tops 10% of revenue.

108
00:06:22,510 --> 00:06:26,769
Services revenue growth must accelerate to offset the new R&D load,

109
00:06:26,910 --> 00:06:29,750
or hardware margins erode within two cycles.

110
00:06:30,119 --> 00:06:33,910
On-device models push more processing into custom silicon,

111
00:06:34,140 --> 00:06:37,790
which raises component costs faster than average selling prices.

112
00:06:37,904 --> 00:06:40,255
can climb across every iPhone tier.

113
00:06:40,625 --> 00:06:43,065
TSMC faces allocation pressure

114
00:06:43,304 --> 00:06:46,165
because Apple will claim additional leading edge wafers

115
00:06:46,165 --> 00:06:49,125
to meet iPhone AI volume targets next year.

116
00:06:49,505 --> 00:06:52,424
Q Qualcomm sees an opening to pitch hybrid cloud on

117
00:06:52,424 --> 00:06:56,825
device solutions to Android OEMs during the resulting capacity crunch.

118
00:06:58,019 --> 00:07:01,489
AMD integrates AI into space computing systems.

119
00:07:01,779 --> 00:07:05,769
Radiation hardened AI accelerators change satellite economics

120
00:07:05,769 --> 00:07:08,500
once models run inference without ground links.

121
00:07:08,859 --> 00:07:13,119
Real-time object detection on orbit slashes required downlink bandwidth

122
00:07:13,260 --> 00:07:16,320
by up to 80% for Earth observation missions.

123
00:07:16,704 --> 00:07:20,445
SpaceX must now accelerate its own onboard inference hardware

124
00:07:20,584 --> 00:07:24,404
or accept higher latency on Starlink enabled defense contracts.

125
00:07:24,665 --> 00:07:26,804
Nvidia faces a forced choice between

126
00:07:26,804 --> 00:07:30,065
licensing its architecture into RAD-hard processes

127
00:07:30,265 --> 00:07:35,285
or ceding the entire segment to AMD and its aerospace partners within 2 years.