E129: Sam Altman plays chess with regulators, AI's "nuclear" potential, big pharma bundling & more
2023-05-19 spoken.md · speaker-labeled ▶ watch ← E128 all episodes E130 →
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
Kill dates that landed since E128
1 hit · 0 partial · 1 miss — windows that closed after 2023-05-12 and up to 2023-05-19, auto-scored against price data and never hand-set. verdict · R · α
| idea | verdict | R | α | closed |
|---|---|---|---|---|
| 🌍 Consumer credit bubble bursts into a credit crisis | HIT | +24.1% | +19.9 | 2023-05-13 |
| 🤖 Synthetic biology's Netscape moment | MISS | -59.2% | -62.6 | 2023-05-13 |
Who moved the board
each voice's force on conviction this episode — supports and opposes, weighted exactly as the replay applied them · share = % of this episode's movement
What got argued (15 ideas)
ordered by how hard each idea moved · quotes are verbatim from the transcript, timestamps deep-link into the episode
Friedberg's argument off Mosaic ML's MPT-7B: a competitive foundation model can now be trained on public data for ~$200k, open-source variants proliferate on Hugging Face and outpace the closed labs, and the models keep shrinking until they run on a phone or Apple silicon at the edge of the network. The model layer therefore has no durable moat and no auditable choke point — the winners are the open-weights distributor and the edge-inference silicon, not the closed-model owners. Chamath takes the other side: you still need a thousand-GPU cluster to compile a frontier model and open weights do not transfer to an orthogonal problem without retraining.
So let me just push back on that, because Mosaic ML published this model that is, let me, I can pull up the performance chart, or Nick, maybe you can just find it on their website real quick of the new model they published. Chamath, they trained this model on open source data that's publicly available, and they spent $200,000 on a cluster run to build this model, and look at how it performs compared to some of the top models that are closed source.
The FTC's suit to block Amgen/Horizon ($27.8B) removes the only reliable exit for risky biotech R&D. Chamath's channel check: the crossover funds that hold the public biotech stocks are the same funds financing the private phase-1/phase-2 companies, and if $28B pharma M&A can be stopped they change risk posture and a pall comes over early-stage biotech financing. Sacks generalizes it — IPO and M&A are the only two non-zero outcomes, so taking M&A off the table suppresses already-challenged venture returns. Friedberg opposes: nobody is actually blocking speculative early-stage deals, and Horizon is a mature $4B-revenue business.
It is because the people that traffic in these stocks are the same ones that fund these early stage biotech companies. And I talked to a bunch of them, and they're like, if these guys block this kind of deal, we're going to get out of this game entirely. So just from the horse's mouth, what I'm telling you is you're going to see a pall come over the early stage venture financing landscape because a lot of these guys that are crossover investors, that own a lot of these public biotech stocks, that also fund the private stocks, will change their risk posture if they can't make money.
Look, it is hard enough to make money as either a pharma investor or as a VC that there's only two good outcomes, right? There's IPOs, there's M&A, everything else basically goes, everything else is a zero, it goes bankrupt. So if you take M&A off the table, you really suppress the already challenged returns of venture capital.
Chamath's read on where drug-price enforcement actually lands: follow the dollars and the culprits are the pharmacy benefit managers, not the manufacturers. The FTC issued compulsory orders this week demanding the PBMs turn over their business records, and Chamath calls that the one thing Lina Khan got right — investigate then regulate the middlemen. The tradeable read-through is regulatory and margin risk at the vertically integrated PBM owners rather than at the drug makers.
So to your point, Sacks, the FTC has been worried that what Friedberg said has been happening, but the real sort of middlemen manipulator in this market are the pharmacy benefit managers. And so this week, she actually issued compulsory orders to the PBMs and said, turn over all your business records to me. I'm going to look into them. That makes a ton of sense.
Sacks reading the record-low Gallup 'good time to buy' print (21%): this is not a bottom. Nobody with a 3% mortgage will sell into falling prices and re-borrow at 7%, so existing-home transaction volume and interstate mobility both freeze. The tradeable read-through is the transaction-volume-levered names — brokerages and portals earn per transaction, not per home price — and the frozen market removes the pressure valve that let unhappy people move states.
And then for the new one, you're going to lose your 3% mortgage and have to get a new one at 7%. So you're not going to buy anything like the house you currently have. So it freezes the market. It freezes mobility. I think over the last few years during COVID, you saw tremendous movement between states. I think that's going to slow down a lot now because people just can't afford to trade houses.
And so if that's the trend, then it becomes very hard for a regulatory agency to go in and audit every server or every computer or every computer on the network and say, what model are you running? Is that an approved model? Is that not an approved model? It's almost like having a regulatory agency that has to go in and audit and assess whether a Linux upgrade or some sort of open source platform that's being run on some server is appropriately vetted and checked. And so it's almost like a fool's errand.
I think actually I split the difference here a little bit. I don't think it needs to be an agency and licensing. I do think we have to have a commission, and we do need to have people being thoughtful about those thousand use cases, Chamath, because they are going to cause societal harm or things that we cannot anticipate.
The FTC's claim in the filing, which I actually read and I was like, this is actually a pretty good claim, is that the way that Amgen sets the prices for their pharmaceutical drugs is they go to the insurance companies, the payers and the health systems, and they negotiate drug pricing.
I think it's more of a moat where, because it's not that the ladder comes up and nobody else can get in, but the regulations are going to be a pretty big moat around major incumbents who know they qualify for this because they're going to write these standards.
So we're going to do the same thing here. We're going to buy into these phony narratives here to get the centers off our backs. And that's going to create this giant AI industrial complex that's going to slow down real innovation and be a burden on entrepreneurs.
Listen, I think this choice makes sense on this level. Twitter's business model is advertising. Elon does not like selling advertising. She's really good at selling advertising. He's chosen a CEO to work with who's highly complimentary to him in their skill sets and interests. I think that makes sense. I think there's a lot of logic in that.
So I think we've got not just a huge commercial real estate problem, but it's going to be a big banking problem as basically people stop pretending. Right now they're trying to restructure loans. It's called pretend and extend. You reduce the rate on the loan but add term to it. But that only works for so long. If this keeps going, if the market keeps looking like this, I think we're going to have a real problem. And that will be a problem in the banking system.
Apple's ~$3,000 AR headset (expected to be revealed in June, shipping in the fall) is a deliberate break from Apple's launch-only-when-perfect playbook, and Sacks argues the endgame is glasses with a camera plus AI layered on the world — Terminator mode — which makes it a genuine next compute platform and forces Meta to compete. Chamath opposes on price and utility: you already paid $400 to get notifications on your wrist, so why pay $3,000 to get them on your eyes. Friedberg refuses to engage at all on the grounds that nobody on the pod has seen the product.
Episode digest
written during extraction and stored in data/extractions/ep129.json — the auditable source of truth, including everything market-adjacent that did not earn a capture
Sam Altman's Senate testimony three days earlier is the direct confirmation event for Chamath's `ai-fda-licensing-regime-2023`, and he takes the victory lap — Altman went further than Chamath had sketched, asking for licenses to even compile a model, which Chamath reframes as KYC enforced at the cloud/GPU layer (tax ID to run a big cluster). None of the three opponents moved: Sacks calls it naked regulatory capture and a 2016-disinformation replay that produces an 'AI industrial complex', Friedberg calls model auditing a fool's errand, and Jason still rejects an agency plus licensing while now conceding a commission. The tradeable residue is Sacks conceding the mechanism of `foundation-models-are-a-big-tech-oligopoly-2023` — regulations become 'a pretty big moat around major incumbents' — against a new open-source thesis coined off Friedberg's Mosaic ML MPT-7B receipt ($200k, public data, competitive with closed models, headed for the phone), which Chamath flatly denies. Big pharma delivered the episode's cleanest break: on the FTC's Amgen/Horizon suit, Chamath and Sacks reinforce `antitrust-punitive-not-principled-2022` and warn of a funding chill in early-stage biotech, but Friedberg read the filing and says the FTC's bundling case is compelling — a rare Friedberg defection from the anti-Lina-Khan consensus — while Chamath flags the PBMs as the real drug-price target now under compulsory FTC orders. Sacks brought the hardest receipts of the show on San Francisco office: 35% vacancy, 30M of 90M square feet empty, AI worth only ~1M square feet of demand, zombie towers heading to the banks and 'pretend and extend' running out — plus a not-a-bottom read on housing where 3%-to-7% mortgage lock-in freezes turnover and mobility. Diarization is CLEAN: 16 of 16 addressed-by-name next-turn checks passed, Jason is top talker at 147 turns, timestamps are monotonic with no gaps, and Chamath is unambiguously present and fingerprint-consistent (Social Capital reference, 8090-style trade framing, the mansion price-cut aside) — which supports E128's zero Chamath turns being a genuine absence rather than a merge.