Trendaavat aiheet
#
Bonk Eco continues to show strength amid $USELESS rally
#
Pump.fun to raise $1B token sale, traders speculating on airdrop
#
Boop.Fun leading the way with a new launchpad on Solana.
samsja
leading research at @PrimeIntellect
I am hiring research engineer at @PrimeIntellect
We are building an open source agi labs and are looking for raw talent. We don't care about your previous job title.
Everybody in the research team is full stack, we build infra and also look at data. If you have a sweet spot for system, reinforcement learning, data or scaling law you will be served with ton of challenge to solve
75,82K
text base declarative system will win, it's time to reinvent the computer
17.8. klo 08.09
Maybe llm + nixos will save linux by making their user 10x more powerful by having a computer agi connected to the os. Imagine just asking " can u install cursor and move all my vscode setting to it"
Meanwhile macos user will still be using their mouse to download and configure everything manually
Such a short term vision to want to train a vision model to do click for me, need to redesign everything for agi
2,11K
Maybe llm + nixos will save linux by making their user 10x more powerful by having a computer agi connected to the os. Imagine just asking " can u install cursor and move all my vscode setting to it"
Meanwhile macos user will still be using their mouse to download and configure everything manually
Such a short term vision to want to train a vision model to do click for me, need to redesign everything for agi
17.8. klo 07.46
isn't nixos the end game of sandbox for llm ?
10,54K
RL is so sensitive to numerics, last time torch compile was making some run crash now vllm v1
12.8. klo 11.23
moving from vllm v0 to v1 made our async rl training crash! read how we fixed it
we recently migrated from v0 to v1 as part of a larger refactor of prime-rl to make it easier-to-use, more performant and naturally async. we confirmed correct training dynamics on many smaller-scale runs, but hit a wall when trying to reproduce a larger scale run that ran without problems prior to the refactor. Specifically, training DeepSeek-R1-Distill-Qwen-1.5B on single-turn math problems from our INTELLECT-2 math dataset at 8k context with two-step off-policy delay would crash fatally roughly 400 steps into the training
6,7K
o1/o3 were the real gpt5 and they did delivered hard maybe bigger jump than gpt3 to 4, RL is still continuing to follow scaling law
Pretraining also scale but inference is just too expensive with giant model
Agree tho than open source will win
10.8. klo 12.04
GPT-5 failed twice.
Scaling laws are coming to an end.
Open-source AI will have the Mandate of Heaven.
4,33K
Johtavat
Rankkaus
Suosikit