me: kinda fucked up how many people are offloading their whole lives to muse or instinct or grokbot or whatever
claude: yeah
claude: yeah
77
248
8,830
183,124
like it’s so good i can’t even talk about it. usually i’d say something goofy and mysterious
9
0
27
3,268
the data economy is gonna be insane. for process knowledge alone there's ~1 million SMBs who can sell their data for $100k+ to 1000s of data brokers and 100s of RL env cos and 10s of labs. $100T market easily
30
23
780
49,334
will brown retweeted
We finished the Training Agents series. Six live sessions over six months, from evaluating agents to training them inside real environments. All of it is on the Hugging Face YouTube channel and all of the code is open.
Here's what we did and who made it happen:
1. Agentic re agent evals actually stand, and why benchmark scores don't match what people see in use. With Avijit Ghosh and Nathan Habib (Hugging Face), Arvind Narayanan (Princeton), Pierre Andrews (Meta), J.J. Allaire (UK AI Security Institute) and Mahesh Sathiamoorthy (Bespoke Labs).
2. RL for Agents Workshop Environments, rollouts, reward design and the inference bottlenecks that appear when you move from RL for LLMs to RL for agents. With Lewis Tunstall (Hugging Face), Will Brown (Prime Intellect), Ofir Press (Princeton) and Alex Zhang (MIT CSAIL).
3. Training Agents 1: SFT on agent traces Public coding-agent traces turned into prompt/completion data, a TRL + LoRA fine-tune on Hugging Face Jobs, metrics in Trackio, and an honest look at what the first eval numbers can and cannot tell you. Joined by Sergio Paniego and Quentin Gallouédec.
4. Training Agents 2: Distillation Off-policy, on-policy and self-distillation for moving capability from a teacher into a smaller coding agent.
5. Training Agents 3: Reinforcement learning GRPO after SFT: group sampling, verifiable reward functions, reading the reward/KL/length curves, and three experiments, one of them with a deliberately gameable reward so we could watch the hacking happen.
6. Training Agents 4: From reward functions to environments The reward stops being a function and becomes a place the agent acts in. We walked the reset()/step() contract from Gym to LLM agents, built an OpenEnv environment and pushed it to the Hub, plugged it into TRL's GRPOTrainer, then trained a real coding agent (OpenCode) through Harbor with AsyncGRPOTrainer on Hugging Face sandboxes.
The series has passed 300k views. Thank you to every speaker, to the TRL team, and to everyone who showed up live with questions.
Playlist: youtube.com/playlist?list=…
Here's what we did and who made it happen:
1. Agentic re agent evals actually stand, and why benchmark scores don't match what people see in use. With Avijit Ghosh and Nathan Habib (Hugging Face), Arvind Narayanan (Princeton), Pierre Andrews (Meta), J.J. Allaire (UK AI Security Institute) and Mahesh Sathiamoorthy (Bespoke Labs).
2. RL for Agents Workshop Environments, rollouts, reward design and the inference bottlenecks that appear when you move from RL for LLMs to RL for agents. With Lewis Tunstall (Hugging Face), Will Brown (Prime Intellect), Ofir Press (Princeton) and Alex Zhang (MIT CSAIL).
3. Training Agents 1: SFT on agent traces Public coding-agent traces turned into prompt/completion data, a TRL + LoRA fine-tune on Hugging Face Jobs, metrics in Trackio, and an honest look at what the first eval numbers can and cannot tell you. Joined by Sergio Paniego and Quentin Gallouédec.
4. Training Agents 2: Distillation Off-policy, on-policy and self-distillation for moving capability from a teacher into a smaller coding agent.
5. Training Agents 3: Reinforcement learning GRPO after SFT: group sampling, verifiable reward functions, reading the reward/KL/length curves, and three experiments, one of them with a deliberately gameable reward so we could watch the hacking happen.
6. Training Agents 4: From reward functions to environments The reward stops being a function and becomes a place the agent acts in. We walked the reset()/step() contract from Gym to LLM agents, built an OpenEnv environment and pushed it to the Hub, plugged it into TRL's GRPOTrainer, then trained a real coding agent (OpenCode) through Harbor with AsyncGRPOTrainer on Hugging Face sandboxes.
The series has passed 300k views. Thank you to every speaker, to the TRL team, and to everyone who showed up live with questions.
Playlist: youtube.com/playlist?list=…
52
260
1,914
131,329
will brown retweeted
Happy basically anniversary of the funniest day ever
it is 2am and we are actually for fucking real stuck in a fucking elevator @willcb @1thousandfaces_ @NotTuxedoSam
3
4
197
12,067
once we solve continual learning AGI just imagine the kinds of usersim RL envs we can build
15
3
282
15,270
thesis: haha RL envs after AGI
antithesis: continual learning still needs to learn lol
synthesis: haha “we”
antithesis: continual learning still needs to learn lol
synthesis: haha “we”
2
0
27
3,137
okay but for real though, why aren’t they working on this more? how can you claim to “train models to be aligned with human values” when “what it means for humans and AI to treat each other well” is a stated unknown?
BREAKING: anthropic announces verified solutions to 482 open problems in philosophy in advance of IPO
4
5
65
6,824
BREAKING: anthropic announces verified solutions to 482 open problems in philosophy in advance of IPO
250
189
4,681
486,945
"the bitter lesson for search is to grep over a filesystem, just like how google works" - rich sutton
42
45
1,210
59,848
will brown retweeted
Love seeing the open problems in continuous learning laid out this clearly. Line of sight to “owning your intelligence” and reasons to move away from relying on models from the big labs. Not all problems need model weight changes, but lots of important ones do.
> there are major open challenges around knowledge acquisition, harness-weight tradeoffs, user modeling, dynamic underspecified preferences, updating and planning across timescales, selective forgetting, data efficiency, learning from golden answers, and company-scale simulation. this is the roadmap, roughly speaking.
> there are major open challenges around knowledge acquisition, harness-weight tradeoffs, user modeling, dynamic underspecified preferences, updating and planning across timescales, selective forgetting, data efficiency, learning from golden answers, and company-scale simulation. this is the roadmap, roughly speaking.
applied research at prime intellect is both a client-facing team and a fundamental research team. continual learning is our north star. we view this dual approach as necessary in order to study the core challenges faced by models which self-improve in production. the real world is big and messy.
we spend part of our our time working with customers to train SOTA models for specialized domains, and we automate and productize our work as we go. we also do open-ended research on novel methods for self-improvement and long-horizon memory, and we get to experiment with data that no other lab gets to see.
today, the most readily-deployable form of continual learning is converting "exhaust" (agent traces, logs, feedback) into robust environments and tasks for continual RL. but this is not the final form. there are major open challenges around knowledge acquisition, harness-weight tradeoffs, user modeling, dynamic underspecified preferences, updating and planning across timescales, selective forgetting, data efficiency, learning from golden answers, and company-scale simulation. this is the roadmap, roughly speaking.
some of these may require changes to architecture, or parameter-efficient methods, or fundamentally new algorithms. we have our hypotheses and early evidence, but everything is on the table. the world is our oyster.
we are betting on a radically different future of what ai deployment will look like, and we can see the path more clearly every day. it is a world where agents learn like humans do, with billions of individual models compounding their learnings from the real world continuously.
we're hiring incredibly ambitious researchers in SF, NY, and London. come join us.
0
1
13
1,880
will brown retweeted
cool kids of RL, I always learn a ton from what they build 🙌🏼
you gotta join them
you gotta join them
over the past few months we've been rapidly scaling our partnerships with leading companies across a bunch of incredibly exciting domains
we collaborate with their in-house experts to deploy continual learning loops in prod
hiring for many new roles :)
primeintellect.ai/careers
1
4
86
7,095
would not be a bad time for all of those theorem-proving startups to pivot into software especially now that math is over
10
3
206
10,908
will brown retweeted
Prime agent #2 on FrontierMemory ranking after Letta Dream!
Great to see Prime Agent do so well on this new self-improving memory benchmark.
My only nit: Let's add agent versions to results!
Great to see Prime Agent do so well on this new self-improving memory benchmark.
My only nit: Let's add agent versions to results!
FrontierMemory also allows measuring the ability of agent harnesses to improve over time. We find that Letta Dream and @PrimeIntellect's Prime Agent improve the most from learning from prior experience data.
5
6
40
3,038