E166

E166: Mind-blowing AI Video: OpenAI launches Sora + Is Biden too old? Tucker/Putin interview & more

2024-02-16 spoken.md · speaker-labeled ▶ watch ← E165 all episodes E167 →

3
ideas born
11
ideas moved
23
captures · 4 voices
6
dissenting
+179.7
conviction added
-143.1
decay · 116 silent

Every number here is replayed from score_events — the same ledger the pool ranks on. Decay is what the 116 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

▲ +72.5 🤖 Generative video collapses content production cost born at green threshold 72.5 still ember — green gate not met
▲ +18.9 🤖 AI collapses the cost of knowledge work to ten cents on the dollar ember watch 35.9 → 54.8
▲ +50.3 🏛️ Biden stays the nominee - there is no mechanism to remove him born at watch 50.3
▲ +15.5 🤖 Apple makes huge AI gains in 2024 (contrarian) ember watch 34.4 → 49.9
▲ +11.3 📈 Code-for-hire outsourcers displace white-collar labor first dormant ember 10.4 → 21.7
▲ +16.8 📈 Arm's post-earnings rip is a float squeeze, not a repricing born at ember 16.8

Kill dates that landed since E165

0 hit · 1 partial · 1 miss — windows that closed after 2024-02-09 and up to 2024-02-16, auto-scored against price data and never hand-set. verdict · R · α

ideaverdictRαclosed
🤖 Google defends search by doubling TAC — margin traded for share MISS -57.5% -80.7 2024-02-11
🏛️ Nobody's incentives point at de-escalation — the war economy ratchets up PARTIAL +8.2% -15.0 2024-02-11

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
9 captures · 40% of movement · 2 ideas born
+108.6 / -10.8 → net +97.7
Friedberg
Friedberg
7 captures · 38% of movement · 1 idea born
+69.3 / -46.0 → net +23.3
Jason
Jason
4 captures · 13% of movement
+39.0 → net +39.0
Sacks
Sacks
3 captures · 9% of movement
+23.3 / -3.7 → net +19.7

What got argued (11 ideas)

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

NEW MSFT 🤖 Generative video collapses content production cost closed 22 ▲ +72.5 0.0 → 72.5

OpenAI's Sora shows prompt-to-video works end to end, so the cost of producing film, TV and game content falls toward zero and a new AI-generated content layer forms on top of the existing distribution platforms. The besties' investable read: the winners are the technology platforms that package these tools for consumers (plus the compute behind them) and the content buyers whose production costs deflate; Sacks adds that independent film becomes economic again once the cost curve breaks.

plays MSFT ·primary GOOGL NFLX evals 2025-02-16
Jason
Jason support ×2 explicit_prediction ▶ 25:47
if you put something like this out there to the public and said, it's 20 bucks a month to do this, or 50 bucks a month, you can have a lot of consumers who want to buy this product. So when I see this, I think people are going to pay for this.
Chamath
Chamath support ×2 sentiment ▶ 39:47
This is going to revolutionize pornography. I think it's the first place. No, no, I'm not saying it as a joke. I think that's where you're going to see this first
Sacks
Sacks support ×2 explicit_prediction ▶ 42:18
But this is going to be revolutionary for independent film. ... This is going to revolutionize it because it will bring the cost down close to zero.
Friedberg
Friedberg support ×3 explicit_prediction ▶ 50:56
There's going to be this whole new model of AI generated content. The real winners are going to be the technology platforms that bring these tools in a simple, intuitive way
NEW DJT 🏛️ Biden stays the nominee - there is no mechanism to remove him closed 0 CONTESTED ▲ +50.3 -0.0 → 50.3

Sacks' and Chamath's call after the Hur report and the skipped cognitive test: the Democratic Party has no mechanism to replace a candidate who has locked up the primary delegates against his will, the elders set the calendar so Biden clears it by the end of March, and his staff are the last people who would push him out. So Biden is on the ballot in November barring an act of God, with Trump and RFK Jr the other two names. Friedberg takes the other side and gives a switch a real chance. The listed expression is the Trump-linked vehicle, which trades as a proxy on his 2024 odds against a visibly weakened Democratic ticket.

plays DJT ·primary evals 2024-11-16
Friedberg
Friedberg oppose ×2 explicit_prediction ▶ 1:25:47
people are not happy in the party. They're very, like very much unhappy and everyone would love to see someone else ... it's a little late now, but I still think that there's a chance
Sacks
Sacks support ×3 explicit_prediction 9mo horizon ▶ 1:26:50
there is no mechanism to replace a candidate who has won all the necessary primary votes against his will ... He's going to be the candidate, you know, barring act of God
Chamath
Chamath support ×3 explicit_prediction 9mo horizon ▶ 1:30:46
He's running, Donald Trump is running, and RFK is running. Those will be the three people on the ballot in November of 2024
MSFT 🤖 AI collapses the cost of knowledge work to ten cents on the dollar closed 28 CONTESTED ▲ +18.9 35.9 → 54.8
Chamath
Chamath support ×3 explicit_prediction ▶ 34:56
if you look at the amount of OPEX inside of any company that sits inside of R&D, if you look at the amount of stock-based compensation, right, if you replace humans with bots, which is what this is, you're going to crush those things. SBC at the limit goes to zero
Jason
Jason support ×2 explicit_prediction ▶ 59:12
we're going to automate the rest of the jobs with AI and we're going to outsource and have hourly workers or full-time contractors outside the United States. It's the undeniable trend.
NEW ARM 📈 Arm's post-earnings rip is a float squeeze, not a repricing closed 2 ▲ +16.8 0.0 → 16.8

Chamath's structural read on Arm's move after earnings: only ~10-15m of ~150m shares actually float because SoftBank holds ~90%, and about 12.5m shares were short going into the print, so nearly a full turn of the float was short. The move therefore looks like an enormous short squeeze by funds that were long other chip names and shorting Arm as a spread trade to damp volatility, not a fundamental repricing of a flat licensing business - and a float that thin cuts violently in both directions.

plays ARM ·primary SFTBY evals 2025-02-16
Chamath
Chamath support ×2 sentiment ▶ 5:51
there was almost an entire turn of the float that was short. So I don't know how much of this reaction was just an enormous short squeeze of people that were largely long other chip companies.
AAPL 🤖 Apple makes huge AI gains in 2024 (contrarian) closed 0 CONTESTED ▲ +15.5 34.4 → 49.9
Jason
Jason support ×2 explicit_prediction ▶ 36:25
you're kind of going to see, I think, in iOS 18 or the next iPhone 16, they're going to have an image model baked onto that phone
Chamath
Chamath oppose ×2 sentiment ▶ 36:44
Where is Apple in all of this? It's unbelievable.
Friedberg
Friedberg support ×2 explicit_prediction ▶ 36:55
I think you're going to see a chip on the next set of mobile devices by Apple that will run models locally.
SFTBY 📈 SoftBank is a forced seller — the Arm IPO is balance-sheet deleveraging closed 0 CONTESTED ▼ -14.1 14.1 → 0.0
Friedberg
Friedberg oppose ×2 sentiment ▶ 5:31
we may be seeing Masa with his whole AI vision story that he's been telling for seven years, we may see Masa clawing his way out of the hole
Sacks
Sacks oppose ×1 sentiment ▶ 8:36
he's a gambler and he is playing a power loss. So he only needs one really big return to basically make up for all the mistakes. And it seems like he's hit that once again.
ACN 📈 Code-for-hire outsourcers displace white-collar labor first closed 7 CONTESTED ▲ +11.3 10.4 → 21.7
Chamath
Chamath support ×2 explicit_prediction ▶ 33:37
this paper essentially now allows you to traverse a code base, understand it, generate these unit tests with high accuracy
NVDA 🤖 Inference cost collapse deflates OpenAI, LLM startups and Nvidia's multiple closed 30 ▼ -10.5 63.4 → 52.9
Chamath
Chamath support ×2 explicit_prediction ▶ 14:20
you're about to see a one-tenthing of the compute cost ... So today, you raise $100 million because you need to spend 70 or 80 on compute. Tomorrow when you only need to spend seven or eight for the same amount of throughput
Friedberg
Friedberg oppose ×3 explicit_prediction ▶ 27:47
when you can do more with less, it doesn't mean you need less chips. It means you need a lot more chips ... it really starts to rationalize the Nvidia thesis that it could be worth ten trillion dollars potentially
META 🤖 Open-source models commoditize the model layer and move to the edge closed 83 ▲ +8.1 50.1 → 58.2
Friedberg
Friedberg support ×2 explicit_prediction ▶ 27:47
these models are getting chopped up in a way that you can run small models in a highly trained way locally, and you don't need a massive general purpose model
DJT 🏛️ Third-party candidates take a Perot-plus share of the 2024 vote, at Biden's expense closed 16 CONTESTED ▲ +6.4 52.4 → 58.8
Jason
Jason support ×2 sentiment ▶ 1:16:02
Third party is seeming more and more viable as we've talked about here. People forget the case of Ross Perot getting 19 percent, RFK looking pretty favorable
Friedberg
Friedberg oppose ×2 sentiment ▶ 1:22:27
Third party candidate can't win in the system.
Chamath
Chamath support ×3 explicit_prediction ▶ 1:28:18
there is a very reasonable chance, and I would put reasonable somewhere between 30 and 50%, that this third party gets created after the presidential election ... if RFK can get 30 plus percent support, which in some states I think you will
AAPL 🤖 Models commoditize — proprietary data is the only AI moat closed 4 CONTESTED ▲ +4.6 66.4 → 70.9
Chamath
Chamath support ×2 sentiment ▶ 29:47
eventually when you're scouring the open Internet, it doesn't matter how fast or how big you scour the open Internet. The open Internet is a bounded space. And so you get to the same answer.

Episode digest

written during extraction and stored in data/extractions/ep166.json — the auditable source of truth, including everything market-adjacent that did not earn a capture

Sora dropped mid-tape and dominated the episode: Friedberg argues generative video is a whole new content paradigm where the platforms that package the tools win, Sacks says it revolutionizes independent film by taking production cost to near zero (while insisting most people still want curation, not personalized media), Jason says consumers will pay $20-50/month, and Chamath says the first real market is synthetic porn - captured as a new idea (MSFT/GOOGL/NFLX). The sharpest tradeable disagreement is on compute: Chamath re-ups his own 2024 annual prediction with a flat 'one-tenthing of the compute cost' and separately argues bigger context windows and parameter counts are marketing-ware with sublinear quality gains, while Friedberg takes the Jevons side - efficiency means you need MORE chips - and says it 'rationalizes the Nvidia thesis' at a ten-trillion-dollar market cap. Apple's AI board got three votes on Jason's own contrarian annual prediction: Jason (image model in iOS 18/iPhone 16) and Friedberg (a local-inference chip in the next devices) support it, Chamath dismisses it ('Where is Apple in all of this?', and Apple's MGIE looks like 'GPT 0.1'). On SoftBank, Friedberg's $125B Arm mark and 'Masa clawing his way out of the hole' cuts against Chamath's forced-seller thesis, but Chamath's own read is that the Arm rip was a squeeze of nearly a full turn of a ~10-15m share float, coined as a new bearish ARM idea. The Biden-age segment produced a genuinely dated, falsifiable political call - Sacks: 'no mechanism to replace a candidate' and Chamath naming Biden/Trump/RFK as the three November names, against Friedberg giving a switch 'a chance' - coined against DJT with a 9-month window. Not captured: the Putin/NATO segment (Sacks' Kennan/Burns analysis is retrospective, and war-economy-defense-spending-ratchet-2023 closed 2024-02-11, five days before air), the Bolt option-loan tax mechanics (no tradeable direction), and Friedberg's speculation that Sora was trained on synthetic Unreal Engine footage, which cuts against the training-data-owners thesis but is a claim about method, not economics.