Sam vs Elon Trial, Coinbase Cuts 14% in AI Pivot, Digesting Meta Earnings | Diet TBPN

Oakland Courtroom Drama, Pocket Sausages Included

TBPN opens on week two of Musk vs. OpenAI in Oakland and immediately sets the tone with Mike Isaac’s courtroom color: rain, hard seats, a butt pillow, five beers the night before, and a lunch of banana, mutant orange, black coffee, and "pocket sausages for travelers." The hosts are half-following the case, half-roasting the survival strategy, which is very much the point: this trial has become elite tech gossip with a stenographer.

Greg Brockman’s Testimony Gets Personal

The meaty part is Brockman’s turn on the stand. He describes his private "journal" — maybe diary, maybe Google Doc, maybe Notes app — as stream-of-consciousness writing that was never meant to be public, and says it’s painful to see it entered into evidence even though he’s not ashamed of it. Then comes the bombshell: Brockman recounts a 2017 equity confrontation where Musk rejected an even split, got visibly angry, and "stormed around" the table so aggressively that Brockman thought he might get hit.

The Equity Fight and the ‘Model 3’ Line

The hosts zero in on Ilya Sutskever’s 2017 email pushing back on Musk’s proposed control structure, especially the killer line asking whether a free Model 3 should make them accept "massively unfavorable terms." It’s a funny detail with a serious subtext: free cars are nice, but the question was who would control what they thought might become the most important technology in history. Brockman’s broader argument is blunt — Musk knows rockets and cars, but "did not and does not know AI."

CASEY, the Commerce Department, and the Invisible AI Review Pipeline

From courtroom drama, they pivot to a more wonky surprise: Google, Microsoft, and xAI have now joined OpenAI and Anthropic in pre-release model screening agreements with the Commerce Department’s Center for AI Standards and Innovation, nicknamed CASEY. The hosts are less shocked by the policy than by how invisible it’s been; the center says it has already completed 40-plus evaluations, including on unreleased models, yet almost nobody in AI seems to talk about the reports.

Why Model Approval Could Become a Bottleneck

This leads to the bigger question: if government review slows public launch, do labs just keep building privately? Pulling in Andrew Curran’s read, the show sketches a weird future where GPT-6 could be stuck in approval while OpenAI is already training 6.1 and 6.2 internally, creating a large gap between frontier capability and what the public can access. The hosts also worry about startup pain, regulatory capture, and a world where building a serious AI company means hiring lobbyists before researchers.

George Hotz’s Open-Source Rebellion

TBPN gives airtime to George Hotz and Tiny Corp’s much more anti-authoritarian stance: don’t gate models, don’t restrict downloads, don’t turn AI into a permit system. They joke about future lawsuits against "the grandma who downloaded Qwen," but the underlying concern is real — if U.S. labs get slowed while Chinese open-weight models keep shipping, public-facing American leadership could erode even if private U.S. capabilities remain ahead.

Coinbase Cuts 14% and Calls It Market Reality Plus AI

Next up is Brian Armstrong’s email announcing a 14% workforce reduction at Coinbase. The hosts stress that Armstrong listed the crypto downturn first and AI-driven productivity second, so this isn’t cleanly an "AI replaced the workers" story; it’s a cyclical business tightening up while betting software teams can do more with agents. The most striking detail is organizational: managers may now be expected to have 15-plus direct reports, and Coinbase is reportedly testing "one-person teams" where one person acts as engineer, PM, and designer with AI support.

Meta’s Earnings: Great Business, Nervous Market

The final major segment is Meta, where the ad machine still looks fantastic — core ad revenue up 33% in Q1 — but investors are fixated on Zuckerberg’s $125 billion to $145 billion capex plan. The hosts frame the market’s fear simply: Meta could be repeating the metaverse playbook, except this time the spend is on GPUs and data centers without the safety valve of a cloud business to monetize spare capacity. That tension sits next to one undeniable fact: Meta remains one of the best businesses on earth, and yet the stock still sold off.

AI Slop Books and a Quick Round of Deals

They close with a few speed-round items: Amazon ebook releases have exploded from about 100,000 per month pre-AI to over 300,000 now, which they dub the rise of "AI slop books." Then come deal notes — Long Lake taking AMEX GBT private for about $6.3 billion, and Mistral crossing $500 million with new investors including BlackRock, Wellington, Nvidia, and even Jamie Foxx — ending on the show’s usual mix of genuine signal and amused disbelief.