NirSD
@nirsd
Followers
1K
Following
11K
Media
542
Statuses
10K
Bio/physics/time series foundation models at 8sleep. Writing at https://t.co/uxSO9lZV7q 🦾==💪
Tartarus
Joined June 2009
Hiring ML/AI engineers, researchers at multiple levels. We are building the intelligence layer (hardware+software+AI) to improve sleep and health @Eightsleep. If you are a strong programmer with science/ML experience and want to build the future of personalized health, dm me.
1
2
9
Anthropic sent him a cease and desist letter, OpenAI sent him an offer letter.
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our
1
0
3
Beautiful thinking through a fundamental problem in latent representations. This is why I'm on X.
0
0
2
Hyperscalers won't capture all compute. Economies of scale only works if costs are bounded and scale sublinear. Then a linear/flat revenue model is profitable. But token consumption is a power law. Prolific token user can consume oom more tokens but they can't pay oom more.
0
0
1
I am sorry OAI folks, Claude is the better coding agent. GPT5.3 feels quite stupid by comparison.
0
0
0
The most efficient Tic Tac Toe agent I could find online takes ~10,000 game iterations to converge to an optimal policy. The game only has about 6k unique game states! That's way too inefficient.
0
0
0
Learning RL by doing. Where should I start? When in doubt, simplify. The simplest game I could think of? Tic Tac Toe. TicTacToe has an optimal strategy that guarantees you cannot lose. So I'l teach an agent to play TicTacToe. Surely that should be easy right?
0
0
1
This should be really easy right now. Eventually the alpha will get competed away, but as of right now, it is available.
0
0
0
Seems sad but, I don't think this is going to change. Since the calculator usage became widespread, the average student probably got worse every year at mental arithmetic and the trend never reversed.
Ooph. Using AI, people did much better on practice tests, and notably worse on actual exams, when the AI was unavailable. AI might not be good for actually learning yet!
1
0
1
I haven't seen any post that talks about using LLMs for ML research/data work so I wrote one. https://t.co/idS1kYB7j1
0
0
0
Chatgpt will serve ads and collect 4% of any sales they generate. This is a pivotal moment in the evolution of AI.
@sytaylor @FredaDuan This is ChatGPT charging 4% and we collect the fees on their behalf. Everyone gets a free trial that starts after the first sales. Not saying that’s good or bad. Ads definitely cost more for most.
1
0
2
Liberals in San Francisco jump and stomp on a Waymo vehicle that was in mid-transport taking a passenger to their destination:
0
0
2
We need to very quickly build 100 data centers and terawatts of energy supply and storage. We ain't going back to typing code.
Good news guys, Sonnet 4.5 would only be like $500 a month when the VC bucks dry up. Opus would only cost $3000-$15,000 bucks a month, unless you use Ralph and then it'll be way more. Keep in mind this is one instance. You all starting to see the problem yet?
0
0
2
Statistics was terminally ill. Thankfully ML euthanized statistics. Why did it take so long?
0
0
1