E106: SBF's media strategy, FTX culpability, ChatGPT, SaaS slowdown & more
2022-12-03 spoken.md · speaker-labeled ▶ watch ← E105 all episodes E107 →
Every number here is replayed from score_events — the same ledger the pool ranks on. Decay is what the 93 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
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 (11 ideas)
ordered by how hard each idea moved · quotes are verbatim from the transcript, timestamps deep-link into the episode
Chamath's investable conclusion from ChatGPT: every chat model sitting on the same public substrate converges, so the models themselves get commoditized and the durable advantage is a proprietary training corpus nobody else can license. The winning move is vertical integration to own the data — buy the hospital system to get the breast-imaging corpus, use consumer devices as Trojan horses to accumulate training data (his Apple Watch/ECG example: 'That is the oil') — and to pick end markets where a regulatory pathway to deploy the model already exists.
Friedberg's read three days after ChatGPT launched: the interesting thing is not the novelty, it is that a natural-language chat interface constructs the answer instead of returning ten blue links, which turns Google's OneBox into the whole product and invites a set of competitors to search itself. Google's core search franchise — and the index-and-link economics under it — can be radically disrupted by chat-native systems, with DeepMind's unknown internal state the main hedge.
Friedberg's dated market prediction three days after ChatGPT launched: a hundred thousand startups emerge, all VC and investor attention rotates to this capability, and the next hype/bubble cycle in Silicon Valley will absolutely be generative AI — an internet-moment reflex where 'this changes everything' is the consensus, so a bubble forms. Capital floods the compute and model layer regardless of whether the applications work.
Chamath's framing: enterprise software becomes single-purpose models that solve a function rather than seat-licensed applications, so 'you can name a bunch of SaaS companies that were purveyors of SaaS that will get replaced by essentially GPT-3 or some other language model' — expense management, forecasting and support first, incrementally. He calls this phase one, before multimodal models and the proprietary-data arms race. Bearish the application-layer SaaS complex, with the value migrating to whoever supplies the models.
The first is that I think we're going to replace SaaS with what I call MAS, which is Models as a Service. And so a lot of what software will be, particularly in the enterprise, will get replaced with a single-use model that allows you to solve a function. So these chat examples are one, and you can name a bunch of SaaS companies that were purveyors of SaaS that will get replaced by essentially GPT-3 or some other language model.
The idea that we want to have completely free, unregulated, Bahamian-based trading environments that we can supposedly trust because someone puts on a good face when there is no real regulatory body and regulatory authority overseeing it, at some point it was going to happen.
Sacks' call off Salesforce's quarter, which he treats as the SaaS bellwether: net new ARR fell two-thirds quarter-on-quarter on flat sales and marketing spend, blowing CAC payback from ~30 months to 155 months. The mechanism is seat-based pricing meeting frozen headcount and layoffs, so the decade-long tailwind of starting each year at 120-150% of last year's revenue from existing customers inverts to 80-90% — top line corrects after valuations already did, sales and marketing headcount gets cut across the industry, and '2X is the new 3X' for the next year or so.
We always think that it's a linear line and that it's super dogmatic and fixed, but there's certainly responsiveness. And the release of the lockdowns in Guangzhou and Beijing this week seems to have been a pretty good indication that when things do get, when the tides do change, leadership there seems to respond, not always, but enough to kind of keep things going.
Episode digest
written during extraction and stored in data/extractions/ep106.json — the auditable source of truth, including everything market-adjacent that did not earn a capture
Taped three days after ChatGPT launched, and this is the besties' first reaction to it — four distinct new theses in one segment. Friedberg's are the search box getting disrupted (Google's core franchise beaten by a chat interface that constructs the answer instead of linking out) plus an explicit market call that the next Silicon Valley bubble cycle 'will absolutely be this generative AI business concept'; Chamath's are SaaS being replaced by MaaS (Models as a Service) and the claim that models themselves commoditize so proprietary data is the only moat — buy the hospital system, use devices as Trojan horses for training data. The honest counterweight: Chamath brushes ChatGPT off as 'a cute toy' whose last one-to-two percent of accuracy is exceptionally hard, Jason agrees 'these changes tend to be slow', and both trash Level 5 autonomy as decades-to-a-century away because there is no regulatory pathway — all captured at their true stance. Sacks delivers the sharpest scoreable number of the episode on SaaS: Salesforce's net new ARR down two-thirds quarter-on-quarter blew CAC payback to 155 months, top line now correcting after valuations already did, with seat contraction replacing seat expansion for the next year — which reinforces the E084 earnings-reset idea and cuts against his own E085 'scale winners escape' corollary. The FTX hour is almost entirely media-culpability argument rather than market calls, though Coinbase gets named three times as the regulated survivor and Chamath declares 'the Chinese growth engine is coming back' as lockdowns lift, opposing E088's China-breaks thesis. Diarization is CLEAN (verified by top-talker, addressed-by-name and fingerprint tests); two single-line interjections are misattributed inside otherwise-correct turns (Chamath's 'Jason, let's be honest' sits in Jason's 1:00:45 turn, Jason's 'Help me out here, Chamath' sits in Chamath's 1:39:35 turn) and neither affects a capture — but note that 58:46-1:00:13 is a replayed archival clip from E061, so nothing in that window was captured as an E106 statement.