Benchmark's Future, ARC-AGI, SpaceX IPO, Epic Games Layoffs, Meta Aims for $9 Trillion, RIP Sora
TL;DR
TBPN opens with a live question nobody in venture wants to touch: can founders ever forgive Benchmark? — They frame the Uber/Travis Kalanick saga as a Ship of Theseus problem, noting only Peter Fenton and Eric Vishria remain from Benchmark’s 2017 six-partner roster while newer partners like Chetan Puttagunta, Everett Randle, and Jack Altman inherited the stigma.
Sora the app is dead, but the bigger point is that AI video didn’t become the instant entertainment singularity people feared. — The hosts argue OpenAI is wisely consolidating into ChatGPT, while pointing to Google’s expensive, rate-limited video products and niche successes like an AI-generated “Love Island”-style TikTok series as evidence the market is real but far from one-shot dominance.
ARC-AGI 3 is the newest hard benchmark for agentic intelligence, and frontier models are basically failing it. — Mike from Arc Prize says the new benchmark has 100+ game-like environments and nearly 1,000 levels, with humans scoring 100% and the best AI systems still under 1%, making it one of the few unsaturated tests left for real exploration, planning, and on-the-fly world modeling.
SpaceX IPO rumors instantly moved the entire public-space basket, even before anything is official. — Katie Roof’s scoop sent stocks like AST SpaceMobile up 7.8%, BlackSky up 4.8%, and Intuitive Machines up 4.1%, while the hosts joked the real treasure in an S-1 would be broken-out xAI economics, especially around inference pricing and model profitability.
The Meta-Manus situation suddenly looks a lot more geopolitical than the ‘Singapore company’ story suggested. — The hosts react to reports that Manus co-founders are still in China, have been called in by authorities, and are blocked from leaving while the sale is reviewed, undercutting prior claims that the company was effectively outside Chinese influence.
Nathan Benaich’s core investing advice is brutally simple: don’t build where the smartest AI researchers already want to work. — The Air Street Capital founder, now running a $232 million fund after starting at $27 million, says to avoid obvious collision zones like coding and AI for science, and instead go after domains where workflow knowledge, taste, and distribution matter more than raw model progress.
The Breakdown
The Benchmark forgiveness debate starts with Uber’s shadow
The show opens in full steel-helmet mode: can anyone ever forgive Benchmark for helping force Travis Kalanick out of Uber? The hosts revisit the argument that Uber at roughly $150 billion today looks like a muted outcome compared with the “trillion-dollar” alternate history people imagine under Travis, especially if Uber had kept building its own self-driving stack instead of killing it.
Ship of Theseus, but make it venture capital
They turn the whole thing into a philosophical riddle: if most of Benchmark’s partnership has turned over, is it still the same firm that made the Uber decision? In 2017 the equal GPs were Bill Gurley, Eric Vishria, Matt Cohler, Mitch Lasky, Peter Fenton, and Sarah Tavel; today only Peter and Eric remain from that era, which lets the hosts ask whether founders should eventually treat Benchmark as a different ship.
RIP Sora, and why AI video still feels oddly early
OpenAI is winding down the standalone Sora app, and the hosts actually read that as a healthy sign of product discipline rather than failure. Their bigger point is that Sora and Meta’s video launches were met with “this will one-shot humanity” vibes, yet in practice AI video has behaved more like a cool creative tool than an irresistible new network, especially when creators can just post output to TikTok or Instagram.
Compute, rate limits, and the economics that kill retention
The conversation gets tactical fast: even Google’s video product was expensive, heavily rate-limited, and hard to use at scale. They make the simple but sticky analogy that if TikTok stopped you after five minutes and told you to come back later, it would be dead on arrival — which is basically how a lot of AI video apps feel today.
Fiverr plants an AI-director billboard in Hollywood
One of the more human moments comes from a sighting on the 101: a giant Fiverr billboard telling Hollywood to “Find the best AI directors.” The hosts love the audacity, laugh at KTLA treating AI like it’s causing fender-benders in traffic, and then point out the business flaw underneath it all: if Billy Bowman is a real creator with his own site and Instagram, the ad is basically teaching buyers how to bypass Fiverr entirely.
SpaceX IPO rumor season begins, and the whole space market jumps
A Katie Roof scoop about SpaceX aiming to file for an IPO lit up the retail-space crowd immediately. The hosts joke about a future “Golden Scoop” award while also noting the real analytical gold would be seeing xAI economics in public filings — token pricing, inference margins, and whether any foundation-model business can be read cleanly from a real S-1.
Manus, China, and the deal that suddenly looks harder
Then the tone shifts: reports suggest the Manus co-founders are still in China, have been summoned by authorities, and are being asked not to leave while the Meta sale is reviewed. The hosts are stunned because the earlier public framing was “this is basically a Singapore company,” and they say the whole thing now feels almost predictably like what would happen if a strategically useful AI harness company tried to sell to a U.S. giant without fully exiting first.
ARC-AGI 3 arrives, and Mike from Arc Prize says frontier models are nowhere close
Mike explains ARC’s mission in plain terms: be a public sense-making tool for how close we are to AGI and push labs toward the remaining gaps. ARC-AGI 3 shifts from static puzzles to interactive game-like environments with over 100 games and nearly 1,000 levels, testing exploration, planning, world modeling, and continual learning — and right now, despite all the hype, humans score 100% while the best frontier systems are still under 1%.
Nathan Benaich on the real AI opportunity outside the obvious hype zones
Air Street Capital’s Nathan Benaich closes on a more grounded founder note. He says Europe doesn’t need its own Google, but there are real openings in areas like defense and other sovereignty-sensitive sectors, and his blunt operating advice is memorable: ask where the smartest AI researchers most want to work, then don’t build there — because that’s where you’re most likely to get steamrolled.