Google just turned this MMO into an AGI experiment
TL;DR
Google DeepMind is turning EVE Online into an AI proving ground — Wes and Dylan say DeepMind took a minority stake tied to CCP’s transition and wants to use EVE’s player-driven economy, wars, logistics, betrayals, and real-market-like dynamics as a far messier benchmark than clean lab tests.
Household robots are entering a data flywheel phase, not a polished consumer phase — Figure is reportedly producing one robot per hour, while 1X is sending robots into homes where many tasks are still teleoperated so companies can harvest real-world data and improve fast via firmware.
The near-term robot future is 'robotic slop,' not perfect android maids — the hosts argue the first valuable home use cases are low-dexterity chores like moving clutter, boxing toys, and basic tidying, not cooking dinner or assembling furniture flawlessly.
Quantum is becoming a real privacy threat, not sci-fi garnish — citing Scott Aaronson, they warn that within roughly three years quantum-capable systems could crack older encryption, exposing years of stored government, military, corporate, and personal data that nation-states have quietly archived.
Their biggest business point: compute should be treated more like an asset than an expense — Dylan argues critics calling AI firms 'unprofitable' missed that billions spent on GPUs/TPUs resemble acquiring strategic infrastructure, which is why Larry Fink’s idea of compute as a futures-traded asset class feels plausible.
They think Google may be the sleeper winner of the AI race — not because Gemini alone dominates, but because Google can embed AI across Search, Chrome, Android, YouTube, cloud, Waymo, and stakes in companies like Anthropic and SpaceX while monetizing even long-tail queries better than old search did.
Summary
The robot future starts weird, and fast
They open with a funny but telling image of the future: a house full of semi-autonomous gadgets that feel a little too alive. That sets up the first real point — Figure is ramping much faster than expected, now producing roughly one robot every hour, which they see as the start of a robotics flywheel where shipping imperfect robots matters more than waiting for perfection.
1X, teleoperation, and why early buyers are basically paying to be beta testers
Dylan brings up 1X, saying the company is rolling out home robots around the $20,000 mark, with many tasks still completed by remote human operators. The clever part, he says, is that every teleoperated dish-stacking session or home chore becomes training data in a real environment, which could generalize across homes and turn today's clunky service into tomorrow's autonomy.
'Robotic slop' is the real consumer wedge
One of the more memorable phrases in the episode is the 1X founder's idea of 'robotic slop' — the mountain of low-skill, low-dexterity chores people still hate doing. Wes immediately relates with a painfully human story about a Costco standing desk sitting 90% assembled on his floor for three weeks, which becomes their perfect example of why even mediocre household help could be hugely valuable.
The darker side of home robots: hacks, accidents, and rollout strategy
They pivot from convenience to nightmare fuel pretty quickly: teleoperated robots, hacked household devices, and the first inevitable 'horrific incident.' Their conclusion is that B2B deployment will probably come first — warehouses, hotels, controlled spaces — because the optics and safety risks are much easier to manage than a robot going off-script in someone's home.
Quantum could crack the past open
The conversation then jumps to Scott Aaronson's warning that quantum systems may break much of today's encryption within about three years. The hosts linger on the unsettling implication: nation-states have been storing encrypted traffic for years, and once quantum-resistant systems arrive, all the old intercepted material — from VPN traffic to state communications — could suddenly become readable.
AGI, jobs, and a post-scarcity future that only works if governance does
From there they get philosophical: Mark Andreessen saying 'AGI is here, just not evenly distributed,' whether AI can end world hunger, and what happens when GDP is no longer tightly linked to human labor. Dylan's framing is that losing jobs only feels catastrophic because people don't trust that resources, healthcare, and housing will still be accessible — if that trust existed, mass automation might look more like liberation than collapse.
Why EVE Online is basically catnip for Demis Hassabis
The emotional high point of the episode is DeepMind's EVE Online move. Dylan lights up explaining EVE as one of the nerdiest, deepest games ever made: mining, blueprints, factories, markets, alliances, betrayals, lawless outer systems, and wars with real-world dollar equivalents — plus his own story of grinding for a year to buy a dream ship that got blown up on day one.
Compute as the real asset class — and why Google might win
They close on business strategy: Larry Fink's suggestion that AI compute could become its own futures market, and Dylan's argument that buying compute is more like acquiring productive infrastructure than burning cash. That rolls into a broader thesis that Google may be uniquely well-positioned because it can spread AI across Search, Chrome, Android, YouTube, cloud, and products people already use — even if its rollout, like Gemini Nano silently downloading onto Chrome, feels a little too aggressive.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.