building @spring_silicon prev @harvard now @thielfellowship
Brooklyn, NY
Joined May 2020
- Tweets28
- Following290
- Followers768
- Likes686
Milan Lustig retweeted
Linus Torvalds says tracking down a bug in Linux's Intel GFX Driver was a "debug session from hell." The AI helping him repeatedly called it "impossible and unsolvable" and suggested they just write a report instead.
He kept pushing. 24 debug patches and 18 kernel boots later he found the bug: one round_up() that should have been round_down().
He then let the AI write the commit message.
git.kernel.org/pub/scm/linux…
Milan Lustig retweeted
Latency is a killer in Physical AI.
every robotics company runs inference in the cloud and pretends latency doesn't matter. it matters. a 200ms round trip to AWS means your robot is blind for 200ms while moving at 2 meters per second. that's 40cm of movement with no feedback. in a warehouse full of humans, 40cm is the difference between a near miss and an OSHA investigation. the companies getting this right are running everything on edge, Jetson or custom silicon, and paying 3x the hardware cost to avoid the latency tax. cloud inference for robots is a demo trick, not a deployment strategy.
Milan Lustig retweeted
I’ve heard from engineers at several frontier bimanual and humanoid robot companies that world action models have completely changed the game and “solved manipulation”. The bottleneck seems to be shifting to running these models at reasonably high frequencies for real-time execution
Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:
• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours,
• this human data scaling law implied a scaling law on never seen robot data,
• both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge
🧵
Milan Lustig retweeted
The best part about this rotation from software into industrialization
Is that venture capital dollars and companies will finally be transforming the physical world
The 2010s deployed billions of dollars, but if you stepped out on the street, the world looked the exact same...
the weather in sf is too nice you need some classic east coast hell (95 and sunny -> 60 and rainy within a day) for motivation
why are there no accurate and public benchmarks for max FLOPS by precision on almost every piece of consumer / edge ml hardware
Why so few late night food options in sf - how is one supposed to lock in without 1am chicken over rice
Milan Lustig retweeted
Thiel Fellow and cofounder of Opt32 @highpriestofswo on what's happening with CS majors:
"I think there's going to need to be a reallocation of talent and focus within the field of CS."
"As we see artificial intelligence capabilities advance, we see a move up layers of abstraction. To where skills like system architecting and higher-level theory are going to become more and more important, and very low-level implementation details — like being a front-end developer — don't provide as much value.
"I don't necessarily think CS as a whole is dropping off, so much as it's reallocating toward those higher levels of abstraction."
Milan Lustig retweeted
Welcome 2026 Thiel Fellows!
WHO ARE THEY?
Victor Boyd: Birmingham, AL - @VictorWBoyd
Cavalla is on a mission to get anything anywhere in under 5 hours. Starting by building autonomous forklifts, through to developing hypersonic highways.
Samuel Carvalho: Recife, Brazil - @samuelclcc
Praso is building the new infrastructure for wholesale commerce — powering procurement, credit, and workflow tools for SMBs across underserved areas in Brazil.
Nick Dobroshinsky: Sammamish, WA - @NDobroshinsky
EveryTicker is democratizing institutional-grade financial research across the entire U.S. stock market, including the thousands of smaller companies Wall Street ignores.
Ishan Gupta: Kanpur, India - @ishangpta
Juicebox is building an AI recruiter that helps companies make better hiring decisions. Agents that understand real skills and move hiring from guesswork to true meritocracy.
Antoni Kiszka: Strzyżowice, Poland - @antoni_kiszka
Derpetual is building the infrastructure to create a market for any asset — with leverage.
Milan Lustig: Cold Spring Harbor, NY - @HighPriestOfSWO
Opt32 is building modern compute infrastructure to put AI onboard objects in the physical world — from robots to cars and drones.
Galen Mead: Chapel Hill, NC - @g413n
Standard Intelligence is building aligned general learners, pretraining large models to actively explore and learn from the Internet.
Aubrey Niederhoffer: New York, NY - @needaubrey
Swoop is building the super app for Africa, starting with food delivery in Nigeria and expanding into financial services across the continent.
Harry O'Connor: Cork, Ireland - @HarryOC493
Sentient Machines is a research lab building foundational models for robotics that generalize across tasks and environments.
Alex Shieh: Salem, NH - @alexkshieh
The Antifraud Company is a fraud bounty hunter defending American taxpayers with AI and investigative journalism.
Claire Wang: Los Angeles, CA - @clairebookworm
Claire is building biologically accurate simulations of entire nervous systems, starting with C. elegans. Developing a simulated brain that researchers can communicate with helps lay the foundation for brain-computer interface (BCI) technology.
Kyler Wang: Portland, OR - @kylerywang
Action is an artificial intelligence company in stealth.
Milan Lustig retweeted
@AnthropicAI plz, can u build proper latex rendering into the web interface, ts is killing me
Milan Lustig retweeted
Triton, cuteDSL, cuTile, Helion, TLX, Gluon, TileLang, ThunderKittens, Mojo etc...
