noodling around @googledeepmind, math & cs @columbia. prev @langchain & @pika_labs 🐣

Joined August 2024
48 Photos and videos
Pinned Post
how you do something is how you do everything this morning I met an insanely cool researcher who craves complexity in everything. so much so that he'd rather get lost in Mongolia or Namibia (unfamiliar, unmapped) than vacatian in Italy. Italy's too easy and boring to him. he has zero interest in easy. i really believe your competitive edge is not a skill you can do at work that no one else can. it's a personality trait that shows up everywhere
8
4
118
11,983
if you’re also curious about robotics hardware and don’t know a lot (like me), i found the best place to start!! this website has an interactive breakdown of literally every component in a humanoid: skeleton, motors, batteries, reducers, sensors, all the way to cost breakdowns and sourcing you can click through real robots (like boston dynamics, apollo) and watch the spec sheets update live also just a joy to use humanityslastmachine.com
day 3, learnt about embodied ai from an incredible blog post bc @FaizHilaly looked me up and down and said I didn’t know enough abt it now i know embodied ai isn’t actually a software race. every humanoid collapses into one bottleneck: the actuator, and the actuator collapses into one material: the magnet. china makes 92% of them the frontier of robotics is being decided by materials science, not software like i thought substack.com/@crosscurrents2…
42
128
1,321
122,211
lots of ā€œAI for scienceā€ experiments test if a model can predict one thing once. real science is a loop: propose, test, fail, adapt, repeat. this paper finally builds that loop for materials discovery and lets an ai agent run the whole thing with no dedicated generative model and no trained ranker, the agent just decided what to try next, on a budget. it beat existing pipelines built for this kind of discovery! this feels like a real preview of ai accelerating science. not just brainstorming ideas, but reasoning about which ones are worth testing arxiv.org/abs/2601.20996
3
11
70
6,467
day 3, learnt about embodied ai from an incredible blog post bc @FaizHilaly looked me up and down and said I didn’t know enough abt it now i know embodied ai isn’t actually a software race. every humanoid collapses into one bottleneck: the actuator, and the actuator collapses into one material: the magnet. china makes 92% of them the frontier of robotics is being decided by materials science, not software like i thought substack.com/@crosscurrents2…
day 2 of this experiment. this evening i realized i didn't know enough about gpu performance, so i read horace he's "making deep learning go brrrr." now i know your gpu spends most of its time waiting for data, not doing math. and that every perf problem is one of three failure modes (compute, memory, overhead), and that most people optimize the wrong one also, 10/10 drawings. see them here: horace.io/brrr_intro.html
4
6
99
129,290
having a hacker mindset isn't about writing code. it's about seeing the underlying mechanics of any system, so you can find shortcuts others can't for example, take video game speedrunners. they see the engine's actual rules, not the ones the designers intended you to follow. they move through a game so fast that 'a normal person can't even understand what is happening on the screen.' a lot of systems (job market, academia) work the same way. they're rigid on the surface, hackable underneath. hence this essay's title: 'how to walk through walls' substack.com/@henrikkarlsson…
3
6
126
6,947
fun reminder to occupy your own life every now and then. instead of hovering a few inches above it, keeping score and categorizing each choice: good decision, bad decision, step forward, step back ā€œconsciousness itself might be the rarest thing in the universe. out of all the silent rocks and distant stars and cold, empty space, the fact that we get to feel anything at all is statistically ridiculous.ā€ yearlyblues.substack.com/p/e…
6
2
62
2,895
new summer goal: learn more about economics and geopolitics in a post-AGI world. i want to spend more time thinking about life after work, AI consciousness debates, what material abundance will do to our brains, and what becomes of our institutions
17
1
124
6,698
day 2 of this experiment. this evening i realized i didn't know enough about gpu performance, so i read horace he's "making deep learning go brrrr." now i know your gpu spends most of its time waiting for data, not doing math. and that every perf problem is one of three failure modes (compute, memory, overhead), and that most people optimize the wrong one also, 10/10 drawings. see them here: horace.io/brrr_intro.html
you are probably one, at most a couple, excellent blog posts away from the frontier of whatever you’re curious about i felt like i didn’t know enough about world models this morning so i read fei fei li’s taxonomy of world models. now i know the field’s actual open problems (3d data scarcer than internet video, multi-object sim absurdly expensive, etc) and why every lab is now racing toward the same architecture this is also my long-winded way of saying you should read it too: open.substack.com/pub/drfeif…
8
8
90
15,543
it’s easy to assume that the future is unknown, that it’s probably bad, and that it’s ours to shape but i couldn’t stop thinking about this essay because it argues the opposite on all three: the future is already written, it’s really good, and we can’t do anything to change it mechanize.work/blog/technolo…
5
30
2,675
such a gem of a blog btw
3
504
no, cake is obviously a frontier risk
no coding, no security, no bio, no chemistry should I ask fable how to bake a cake?
7
3
42
5,351
cool reminder that the way things are is usually just the way they ended up, not the way they were meant to be before this breakthrough, surgeons cut the dura for centuries not because it was necessary, just because it was the only tool anyone had most ā€˜standard practice’ is just an old constraint we haven’t questioned enough
The dura is the brain's armor: a membrane so tough that a surgeon normally cuts through it with a scalpel. For the first time in our clinical trials, we inserted the electrode threads of our implant straight through the dura and into the cortex, keeping the dura intact. Here's how we did it 🧵
1
2
53
11,860
you are probably one, at most a couple, excellent blog posts away from the frontier of whatever you’re curious about i felt like i didn’t know enough about world models this morning so i read fei fei li’s taxonomy of world models. now i know the field’s actual open problems (3d data scarcer than internet video, multi-object sim absurdly expensive, etc) and why every lab is now racing toward the same architecture this is also my long-winded way of saying you should read it too: open.substack.com/pub/drfeif…
2
4
99
13,541
Anika reposted
local gym owner warns people lifting weights at home are on a ā€œvery dangerous pathā€
ANTHROPIC CEO WARNS OPEN-SOURCE AI IS ON A "VERY DANGEROUS PATH"
77
2,776
42,783
849,052
on this plain old monday evening i leave you with my favourite murakami quote — on discipline as a mesmerising experience, a kind of self-hypnosis
1
11
119
2,839
can't wait for more AI that has taste in a specific domain and makes decisions. Odessia cracked it for travel, same thing is inevitable for: > apartment hunting (zillow is a grid that can't hold your actual tradeoffs) > job searching (like, does anyone actually use linkedin for this?) > gift giving (ai that knows your friend, good use-case for multiplayer) etc etc. for all of these, i'm tired of getting a broad surface of knowledge written in LLM-speak. i want an expert opinion
new fav product alert!! if you're also planning a last min july 4 trip like me, @OdessiaTravel is 100x better than just using an LLM. will share how I used it in thread but TLDR: > feels like a friend suggesting activities I'd actually do, not SEO-slop > opinionated & makes decisions (e.g. where to stay) – not an endless brainstorm > searches hundreds of flights, airbnbs & activities in seconds and books them within minutes it's made by the founder of @SonderStays which is probably how it has such good travel taste next stop: Kauai, Hawaii!!
2
10
2,943
Anika reposted
SITUATION EXPLAINED: Post-AGI political economy is the most underrated field of study in the world right now. • The question: assume we have actual superintelligence that can automate labor and act agentically in the real world. Now what does the political economy look like? • Three biggest threats: AI extinction (the Yudkowsky scenario), gradual disempowerment (humans become economically obsolete through normal selection pressures), and Butlerian Jihad (a global degrowth regime where nobody gets AGI) • The people actually working on this: Seb Krier at Google DeepMind, Richard Ngo, the Forethought Institute (Tom Davidson and Will MacAskill), Joe Carlsmith at Anthropic, Nick Bostrom • It's like trying to predict the political economy of the Industrial Revolution before it happened, the best person who tried was Karl Marx, and we know how that went @theojaffee: "Nobody is thinking about this. I'm really surprised by how few people are seriously thinking about this."
1
4
55
8,434
the internet has so many cool things today I stumbled upon a museum of Flash websites. here’s SpaceX and GTA in all their 2002 glory probably in a few years, today’s websites will look just as dated webdesignmuseum.org/flash-we…
3
1
32
2,299
for the curious vibe coder, this is one of the best resources i've found for understanding the future of software and interfaces in a world where ai writes code for you it teaches you how fonts are made and why computers draw curves the way they do – but also about what makes code fast and how the internet actually works. and all the graphics are beautiful makingsoftware.com
14
52
869
37,205