E167

E167: Google's Woke AI disaster, Nvidia smashes earnings (again), Groq's LPU breakthrough & more

2024-02-23 spoken.md · speaker-labeled ▶ watch ← E166 all episodes E168 →

2
ideas born
12
ideas moved
21
captures · 4 voices
4
dissenting
+170.0
conviction added
-139.2
decay · 114 silent

Every number here is replayed from score_events — the same ledger the pool ranks on. Decay is what the 114 ideas nobody mentioned gave up this week; it applies only when an episode is processed.

Tier crossings

conviction thresholds crossed by this episode — 65 / 45 / 15 · ideas born here show where they landed. why 65 isn't always green

▲ +12.3 🤖 Open-source models commoditize the model layer and move to the edge watch green threshold 58.2 → 70.5 still watch — green gate not met
▲ +7.5 🏛️ Speculative-competition antitrust blocks chill tech M&A watch green threshold 60.1 → 67.6 still watch — green gate not met
▼ -9.0 🤖 Models commoditize — proprietary data is the only AI moat green threshold watch 70.9 → 62.0
▲ +59.2 🤖 Gemini's ideological tuning breaks Google's AI product born at watch 59.2
▲ +9.7 🤖 Inference cost is the gate on AI search — 10x too expensive today ember watch 40.2 → 49.9
▲ +16.2 🤖 Nvidia's monopoly margins get competed away by custom and open silicon ember watch 30.4 → 46.6
▲ +41.8 🤖 AI capex outruns app-layer revenue - Nvidia's terminal-value problem born at ember 41.8

Kill dates that landed since E166

0 hit · 0 partial · 3 miss — windows that closed after 2024-02-16 and up to 2024-02-23, auto-scored against price data and never hand-set. verdict · R · α

ideaverdictRαclosed
🏛️ East Palestine forces rail safety re-regulation and operator liability MISS -13.3% -37.8 2024-02-17
🏛️ The AI trust-and-safety layer is the next censorship war MISS -57.9% -82.4 2024-02-17
📈 Zantac carcinogen cover-up becomes a pharma liability reckoning MISS -21.6% -46.1 2024-02-17

Who moved the board

each voice's force on conviction this episodesupports and opposes, weighted exactly as the replay applied them · share = % of this episode's movement

Chamath
Chamath
7 captures · 36% of movement · 1 idea born
+96.1 → net +96.1
Sacks
Sacks
6 captures · 31% of movement · 1 idea born
+48.3 / -33.9 → net +14.4
Friedberg
Friedberg
6 captures · 26% of movement
+55.4 / -12.9 → net +42.5
Jason
Jason
2 captures · 6% of movement
+17.1 → net +17.1

What got argued (12 ideas)

ordered by how hard each idea moved · quotes are verbatim from the transcript, timestamps deep-link into the episode

NEW GOOGL 🤖 Gemini's ideological tuning breaks Google's AI product closed 0 CONTESTED ▲ +59.2 0.0 → 59.2

Gemini refuses or falsifies answers on undisputed facts (black George Washington, 'diverse' founding fathers, no answer on IQ data or on Trump's or Biden's legal cases) because a diversity-and-inclusion tuning layer overrides the model, and Sacks' read is that this is not a rushed-launch accident but an accurate self-portrait of Google's monoculture - so the fix will be to make the bias subtler rather than remove it. The tradeable consequence is product quality rather than politics: an information-interpretation engine that will not answer straight loses consumer trust, users defect to whatever tells them the truth, and Google's culture makes it structurally slow to respond. Bearish Google's consumer AI and search franchise. Distinct from the 2023 trust-and-safety censorship debate, whose window closed six days before this episode aired.

plays GOOGL ·primary evals 2025-02-23
Sacks
Sacks support ×3 explicit_prediction ▶ 54:58
This idea that Gemini turned out this way by accident or because they rushed it out, I don't really believe that. I believe that what happened is Gemini accurately reflects the biases of the people who created it ... I think they're simply going to dial down the bias to be less obvious.
Chamath
Chamath support ×2 explicit_prediction ▶ 59:07
Too many of these judgments, I think, will make this product very poor quality and consumers will just go to the thing that tells it the truth.
Friedberg
Friedberg support ×2 sentiment ▶ 1:14:19
Google could be going down the wrong path here in a way that they will lose users and lose consumers and someone else will be there eagerly to sweep up with a better product.
NEW NVDA 🤖 AI capex outruns app-layer revenue - Nvidia's terminal-value problem closed 0 CONTESTED ▲ +41.8 -0.0 → 41.8

Chamath's arithmetic off Nvidia's $22.1B quarter: at a 30-50% required return plus the costs needed to support it, that spend has to generate roughly $45B of downstream revenue, and nobody on the pod can name who earned it - today's AI apps are proofs of concept and toy demos run in sandboxes, not production code, so the buildout is big tech muscling vendors with idle balance-sheet cash rather than real product demand. Revenue can therefore keep compounding for two or three years while the multiple gets cut, exactly as Cisco's did once value migrated to the application layer; Chamath's ratio test is that a $4-5T Nvidia implies a $100T economy underneath it. Friedberg supplies the Oracle-server and low-end-networking analogues; Sacks takes the direct other side with the dark-fiber precedent - if you build it, the applications eventually get written.

plays NVDA ·primary MSFT SMH evals 2026-08-23
Friedberg
Friedberg support ×2 explicit_prediction ▶ 8:23
The real question ultimately will be, does the initial cost of the infrastructure exceed the ultimate value that's going to be realized on the application layer? ... you could look at Cisco during the early days of the Internet build out, and everyone thought Cisco was the picks and shovels of the Internet, and they were going to make all the values go to Cisco. So we're kind of in that same phase right now with Nvidia
Chamath
Chamath support ×3 explicit_prediction 30mo horizon ▶ 16:11
I think the revenue scale will continue for like the next two or three years probably for Nvidia. But the real question is what is the terminal value? ... People ultimately realized that the value was going to go to other parts of the stack, the application layer.
Sacks
Sacks oppose ×3 explicit_prediction ▶ 24:43
As it turns out, all that fiber eventually got used. The Internet went from, you know, dial up to broadband ... So I think that the history of these things is that the applications eventually get written. They get developed if you build the infrastructure to power them ... we're just at the beginning of a wave that's probably going to last at least a decade
NYT 🏛️ AI's training data gets a copyright bill - citations, links and licences closed 16 CONTESTED ▲ +19.7 20.7 → 40.4
Sacks
Sacks support ×1 sentiment ▶ 1:10:56
I think the better AI models are providing citations now and links. Perplexi actually does a really nice job with this.
Jason
Jason support ×2 explicit_prediction ▶ 1:13:32
I just think the authority at which these LLMs speak is ridiculous. They speak as if they are absolutely 100% certain that this is the crisp, perfect answer ... When it should present it with citations.
NVDA 🤖 Nvidia's monopoly margins get competed away by custom and open silicon closed 42 ▲ +16.2 30.4 → 46.6
Chamath
Chamath support ×3 explicit_prediction ▶ 6:29
in capitalism, when you over-earn for enough of a time, what happens is competitors decide to try to compete away your earnings ... In the case of Nvidia, what you're now starting to see is them over-earn in a very massive way. So the real question is who will step up to try to compete away those profits?
Friedberg
Friedberg support ×2 explicit_prediction ▶ 8:23
So I think Chamath's point is right ... So in networking, a lot of the high-end, high-quality networking companies got beaten up when lower cost solutions came to market later.
Sacks
Sacks oppose ×2 explicit_prediction ▶ 20:38
Right now, their market share is something like 91 percent. That's clearly going to come down, but their moat appears to be substantial. The Wall Street analysts I've been listening to think that in five years, they're still going to have 60-something percent market share.
SPY 🌍 Risk-off overhang lifts - equities and high-beta growth rally closed 11 CONTESTED ▼ -13.5 63.4 → 49.9
Sacks
Sacks oppose ×2 explicit_prediction ▶ 1:17:36
What's happening in the war is that the Russians just took this city of Deikha, which basically totally refutes the whole stalemate narrative, as I've been saying for a while. It's not a stalemate. The Russians are winning ... Why do I think this is a big deal? Because if something like this happens, it could really expand the Ukraine War ... this could lead to a major escalation in the war.
META 🤖 Open-source models commoditize the model layer and move to the edge closed 83 ▲ +12.3 58.2 → 70.5
Jason
Jason support ×2 sentiment ▶ 56:08
People are just not going to want a model that has all this baked in weird bias. They're going to want something that's open source.
Friedberg
Friedberg support ×2 explicit_prediction ▶ 56:22
eventually everyone will find something that they don't want or that they're not expecting, and they're going to say, I don't want to use this product anymore. And so it is actually an opportunity for many models to proliferate, for open source to win.
NYT 🤖 Training-data owners (NYT, Reddit, X, YouTube) win 2024 closed 12 CONTESTED ▲ +11.7 52.0 → 63.7
Chamath
Chamath support ×3 explicit_prediction ▶ 1:06:34
like what Google did with Reddit, we're now going to spend $60 billion a year licensing training data, right? We're going to scale this up by a thousand fold ... we are going to be the truth tellers in this new world of AI
NVDA 🤖 Inference cost is the gate on AI search — 10x too expensive today closed 26 ▲ +9.7 40.2 → 49.9
Chamath
Chamath support ×3 explicit_prediction ▶ 37:20
They are toy apps that are too slow that require too much infrastructure and cost. So the potential is for us to enable that monetization leap forward.
AAPL 🤖 Models commoditize — proprietary data is the only AI moat closed 4 CONTESTED ▼ -9.0 70.9 → 62.0
Chamath
Chamath support ×2 explicit_prediction ▶ 1:14:41
aren't you worried that somebody with an agenda and a balance sheet could now basically gobble up all kinds of training data that make all models crappy? And then they basically put their layer of interpretation on critical information for people?
Friedberg
Friedberg oppose ×2 explicit_prediction ▶ 1:15:29
I think the open Internet has enough data that there isn't going to be a monopoly on information by someone spending money for content from third parties.
NVDA 🤖 Inference cost collapse deflates OpenAI, LLM startups and Nvidia's multiple closed 30 ▲ +7.7 52.9 → 60.5
Chamath
Chamath support ×3 explicit_prediction ▶ 27:55
you now see developers stress testing us and finding that we are meaningfully, meaningfully faster and cheaper than any Nvidia solution, there's the potential here to be really disruptive ... So there's a lot of market cap for Groq to gain by just being able to produce these things at scale
GS 🏛️ Speculative-competition antitrust blocks chill tech M&A closed 53 CONTESTED ▲ +7.5 60.1 → 67.6
Friedberg
Friedberg support ×2 explicit_prediction ▶ 17:18
The other one is that antitrust authorities are blocking all of their acquisitions. And so what do you do with all that cash? ... frankly, we're not going to be able to grow through M&A because of antitrust right now anyway
IRDM 🤖 Pervasive satellite internet covers the whole earth within five years closed 18 ▲ +6.6 28.7 → 35.3
Sacks
Sacks support ×1 sentiment ▶ 34:02
the big moneymaker at SpaceX is Starlink, which is the satellite network, basically broadband from space. And it's on its way to handling, I think, a meaningful percentage of all internet traffic.

Episode digest

written during extraction and stored in data/extractions/ep167.json — the auditable source of truth, including everything market-adjacent that did not earn a capture

Two threads carried the market content. On Nvidia's third straight blowout ($22.1B revenue, +265% y/y, +$247B market cap in a day), Chamath reran his rule-of-capitalism call - over-earning invites competition - and layered on a demand test: $22B of quarterly spend needs ~$45B of downstream revenue, today's AI apps are 'toy apps' run as demos, and Nvidia is Cisco 2000, where revenue compounds while the multiple gets cut as value migrates to the application layer. That is coined as a new idea; Sacks took the direct other side (dark fiber all got used, the applications always get written, a decade-long wave) and, notably, flipped off the merchant-silicon thesis he supported at E143, now arguing GPUs are far harder to commoditize than Cisco's boxes and Nvidia holds 60-something percent share in five years. Chamath's Groq segment (3,000 customers in three days, 'meaningfully faster and cheaper than any Nvidia solution') is a direct double-down on his own E160 inference-cost-collapse prediction and on the inference-cost-gate thesis - flagged as self-interested, he is Groq's seed investor. On Gemini's image scandal, the natural home (`ai-trust-safety-layer-bias-fight-2023`) died 2024-02-17, six days before air, so a fresh bearish-GOOGL idea was coined: Sacks says the bias is a self-portrait of Google's monoculture and will be made subtler rather than removed, Chamath and Friedberg both predict consumer defection. Chamath's fix - spend $60-100B/year licensing training data and be 'the truth teller' for a $10T market cap - lands as strong support on Jason's `training-data-owners-2024`, while Friedberg picked the opposite side of Chamath's own data-moat idea, arguing the open internet cannot be monopolized by paying third parties. Left in the digest as untradeable: the deep-tech-investing philosophy segment, Chamath's Perplexity-chips-away-at-Google aside, and Sacks' Moldova/Transnistria news (captured only as an oppose on the risk-premium-unwind idea, since he named no market).