Dark | Light
# ![@willccbb Avatar](https://lunarcrush.com/gi/w:26/cr:twitter::3064259332.png) @willccbb will brown

will brown posts on X about in the, ai, inference, model the most. They currently have [------] followers and [---] posts still getting attention that total [------] engagements in the last [--] hours.

### Engagements: [------] [#](/creator/twitter::3064259332/interactions)
![Engagements Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::3064259332/c:line/m:interactions.svg)

- [--] Week [-------] -31%
- [--] Month [---------] +82%
- [--] Months [----------] -45%
- [--] Year [----------] +797%

### Mentions: [--] [#](/creator/twitter::3064259332/posts_active)
![Mentions Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::3064259332/c:line/m:posts_active.svg)

- [--] Week [---] +6.90%
- [--] Month [---] +100%
- [--] Months [---] +24%
- [--] Year [-----] +385%

### Followers: [------] [#](/creator/twitter::3064259332/followers)
![Followers Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::3064259332/c:line/m:followers.svg)

- [--] Week [------] +0.77%
- [--] Month [------] +3.50%
- [--] Months [------] +32%
- [--] Year [------] +234%

### CreatorRank: [-------] [#](/creator/twitter::3064259332/influencer_rank)
![CreatorRank Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::3064259332/c:line/m:influencer_rank.svg)

### Social Influence

**Social category influence**
[technology brands](/list/technology-brands)  22.52% [stocks](/list/stocks)  9.91% [finance](/list/finance)  5.41% [social networks](/list/social-networks)  1.8% [products](/list/products)  0.9% [countries](/list/countries)  0.9% [financial services](/list/financial-services)  0.9%

**Social topic influence**
[in the](/topic/in-the) 8.11%, [ai](/topic/ai) 7.21%, [inference](/topic/inference) #235, [model](/topic/model) #1473, [meta](/topic/meta) 5.41%, [agentic](/topic/agentic) #494, [anthropic](/topic/anthropic) #1734, [future](/topic/future) 3.6%, [lab](/topic/lab) #1018, [claude code](/topic/claude-code) 3.6%

**Top accounts mentioned or mentioned by**
[@primeintellect](/creator/undefined) [@xeophon](/creator/undefined) [@maziyarpanahi](/creator/undefined) [@huggingface](/creator/undefined) [@alibabaqwen](/creator/undefined) [@allenai](/creator/undefined) [@arceeai](/creator/undefined) [@meta](/creator/undefined) [@nvidia](/creator/undefined) [@zaiorg](/creator/undefined) [@shielsai](/creator/undefined) [@shiels_ai](/creator/undefined) [@hallerite](/creator/undefined) [@osoleve](/creator/undefined) [@faldufenil12](/creator/undefined) [@teilomillet](/creator/undefined) [@heerprofessor](/creator/undefined) [@rasdani](/creator/undefined) [@hanchunglee](/creator/undefined) [@stalkermustang](/creator/undefined)

**Top assets mentioned**
[Alphabet Inc Class A (GOOGL)](/topic/$googl) [Morgan Stanley (MS)](/topic/morgan-stanley) [DoorDash Inc. (DASH)](/topic/doordash)
### Top Social Posts
Top posts by engagements in the last [--] hours

"your research artifacts need to be Legible to Clankers"  
[X Link](https://x.com/willccbb/status/2021667911560261658)  2026-02-11T19:29Z 39.6K followers, [----] engagements


"the future belongs to those who vaguely understand git"  
[X Link](https://x.com/willccbb/status/2020100043689128199)  2026-02-07T11:39Z 39.6K followers, 272.5K engagements


"how it feels to be deep in the thralls of gpt-5.3-codex psychosis"  
[X Link](https://x.com/willccbb/status/2020290695735701531)  2026-02-08T00:16Z 39.6K followers, 10.1K engagements


"consider ask your employer if your token budget includes training tokens as well as inference tokens"  
[X Link](https://x.com/willccbb/status/2021708138676957534)  2026-02-11T22:09Z 39.6K followers, [----] engagements


"dwarkesh: you should buy more compute dario: were buying a lot dwarkesh: what if its not enough dario: then we make profit. if we buy too much we go bankrupt dwarkesh: but what about the country of geniuses in a datacenter dario: theyll have a nice big datacenter"  
[X Link](https://x.com/willccbb/status/2022414406483886263)  2026-02-13T20:55Z 39.6K followers, 150K engagements


"imo doesnt really make sense to think of it as operating literally at just the token level. the objective is to maximize the relative likelihoods of the rollouts with higher rewards. tokens are just one intermediate representation. mechanistically this looks like identifying the circuit chains whose activations are most correlated with observed rewards. even in the case where all rollouts branch in the first few tokens you have higher-level properties (strategies subsequences) which are more present in the better rollouts and these are captured by latent + sequence-level representations."  
[X Link](https://x.com/willccbb/status/2022874457770238049)  2026-02-15T03:23Z 39.6K followers, [----] engagements


"if you're a later-year PhD student thinking about internships for next summer my team at Morgan Stanley (ML Research) is hiring we're a tight-knit team of [--] researchers working on everything from time-series to RL to diffusion to LLMs both applications and fundamentals. we have compute publishing freedom and an endless supply of hard and interesting problems. feel free to reach out to me with any questions"  
[X Link](https://x.com/willccbb/status/1825261587902423476)  2024-08-18T20:00Z 39.6K followers, 337.8K engagements


"if youre doing a PhD in a quantitative discipline you should print this out and hang it above your desk https://www.lkozma.net/inequalities_cheat_sheet/ineq.pdf https://www.lkozma.net/inequalities_cheat_sheet/ineq.pdf"  
[X Link](https://x.com/anyuser/status/1831677247616397810)  2024-09-05T12:54Z 39.6K followers, 161.2K engagements


"i first got into agentic programming about [--] years ago"  
[X Link](https://x.com/willccbb/status/1868500623718592731)  2024-12-16T03:37Z 39.6K followers, 57.1K engagements


"LLMs know way too many facts. you can ask Llama 1B about the history of pizza and it gives a decent answer. it shouldn't be able to do that. it should just do a google search. use those weights for something else"  
[X Link](https://x.com/willccbb/status/1887489031170109681)  2025-02-06T13:10Z 39.6K followers, 307K engagements


"Of course thats your contention. Youre a first-year AI engineer who learned the hard way that RAG is more than vector search. You just finished reading Building effective agents and were convinced to migrate from LangChain to Model Complex Protocol until next month when"  
[X Link](https://x.com/willccbb/status/1897747876555391415)  2025-03-06T20:35Z 39.6K followers, 140.7K engagements


"coding on your iphone: - acquire mac mini - set up dotfiles nvim etc - install ssh client on phone - ssh to mini - npm install -g @anthropic-ai/claude-code - scripts/tools for web search e2b/docker cli chat - use git branches + aggressively - profit http://CLAUDE.md http://CLAUDE.md"  
[X Link](https://x.com/willccbb/status/1898570734244364598)  2025-03-09T03:05Z 39.6K followers, 181.7K engagements


"11k youtube views on this already is kinda wild to me lol thanks for checkin it out :)"  
[X Link](https://x.com/anyuser/status/1898846931582861689)  2025-03-09T21:22Z 39.6K followers, 133.8K engagements


"the new llama [--] models are so advanced that they require versions of hf transformers that haven't even been invented yet"  
[X Link](https://x.com/willccbb/status/1908606265556877643)  2025-04-05T19:42Z 39.6K followers, 164.9K engagements


"update: i joined @primeintellect :) cannot describe how excited i am to be joining such an incredible team and mission. there is a dire shortage of labs who are truly embracing open-source research. its hard to get the incentives right. you need a business model where open-sourcing your work is positive-sum; being a GPU marketplace is a really good one. prime intellect has been doing incredible work to advance the frontiers of decentralized training and inference and it is only the beginning. my own goal is to continue along the directions of my recent projects and musings but bigger bolder"  
[X Link](https://x.com/willccbb/status/1920538247404663241)  2025-05-08T17:56Z 39.6K followers, 214.6K engagements


"the year is [----]. python [---] now compiles directly to Rython borrow-checked zero-cost no more gc. pip defaults to uv add and tokio ships with stdlib. the GIL has been replaced with lifetimes and del wraps std::mem::forget. we still argue about tabs vs spaces. Today were announcing the preview release of ty an extremely fast type checker and language server for Python written in Rust. In early testing it's 10x 50x even 100x faster than existing type checkers. (We've seen 600x speed-ups over Mypy in some real-world projects.) https://t.co/bAiygYPhqP Today were announcing the preview release of"  
[X Link](https://x.com/willccbb/status/1922340301160538568)  2025-05-13T17:16Z 39.6K followers, 92.8K engagements


"happy saturday for the price of a cup of coffee you can rent a [----] for the day and do some random experiments on small models"  
[X Link](https://x.com/willccbb/status/1923792937042968604)  2025-05-17T17:29Z 39.6K followers, 294K engagements


"being good at ML systems helps you run more experiments. being good at ML theory helps you run less experiments"  
[X Link](https://x.com/willccbb/status/1929364817954927054)  2025-06-02T02:29Z 39.6K followers, 94.9K engagements


"you can just RL a model to play Wordle"  
[X Link](https://x.com/anyuser/status/1929975995475603729)  2025-06-03T18:58Z 39.6K followers, 131.3K engagements


"you can literally make [--] figures for just having a hobby hobbies are just excuses for unambitious people to fill their free time hobbies are just excuses for unambitious people to fill their free time"  
[X Link](https://x.com/anyuser/status/1930063994276196682)  2025-06-04T00:48Z 39.6K followers, 429K engagements


"spending everything on Alexandr Wang"  
[X Link](https://x.com/willccbb/status/1932464175047176277)  2025-06-10T15:45Z 39.6K followers, 120.1K engagements


"2+ years of MCP experience required"  
[X Link](https://x.com/willccbb/status/1937955717081964878)  2025-06-25T19:27Z 39.6K followers, 70.8K engagements


"xAI Head of Product announces that Grok [--] is the Antichrist Going from an office where AI researchers are building the Antichrist to my living room where my girlfriend is watching Love Island is one of the most drastic transitions in the known universe Going from an office where AI researchers are building the Antichrist to my living room where my girlfriend is watching Love Island is one of the most drastic transitions in the known universe"  
[X Link](https://x.com/willccbb/status/1942511578739245534)  2025-07-08T09:10Z 39.6K followers, 204.9K engagements


"and were live been a very long time in the making huge thanks to everyone whos made it possible along the way. cant wait to see what you guys all build here. were just getting started :) Introducing the Environments Hub RL environments are the key bottleneck to the next wave of AI progress but big labs are locking them down We built a community platform for crowdsourcing open environments so anyone can contribute to open-source AGI https://t.co/urAv2hGPCQ Introducing the Environments Hub RL environments are the key bottleneck to the next wave of AI progress but big labs are locking them down"  
[X Link](https://x.com/willccbb/status/1960784880490700959)  2025-08-27T19:21Z 39.5K followers, 295.2K engagements


"were hiring for lots of junior roles right now. apprenticeships basically. can be part-time remote international. if you really want to get our attention for these share a cool env to the hub [---] LOC. well give full-time offers to top performers after. we can sponsor. Called it last year. Junior roles down 23%. Senior roles up 14%. Harvard study tracked [------] firms. Results: Before AI: [--] senior + [--] juniors = [--] person team After AI: [--] senior + Claude = same output https://t.co/wAdQz9Sg8Z Called it last year. Junior roles down 23%. Senior roles up 14%. Harvard study tracked [------] firms."  
[X Link](https://x.com/willccbb/status/1964327414768513259)  2025-09-06T13:58Z 39.6K followers, 419K engagements


"now this is a hiring page Excited to share that I am starting a new company dedicated to the creation of verified superintelligence via autoformalization. Building on the amazing RL infrastructure that we have developed at @morph_labs @mathematics_inc has already achieved a breakthrough result by Excited to share that I am starting a new company dedicated to the creation of verified superintelligence via autoformalization. Building on the amazing RL infrastructure that we have developed at @morph_labs @mathematics_inc has already achieved a breakthrough result by"  
[X Link](https://x.com/willccbb/status/1966204798983971238)  2025-09-11T18:18Z 39.6K followers, 357.9K engagements


"I just read this new paper from Google and Im absolutely buzzing 🤯 The core idea is almost offensively simple: ditch recurrence and convolutions and use only attention. Thats it. And somehowit unlocks a whole new regime of performance scale and simplicity. Heres what blew my mind: - No recurrence full parallelism. Tokens dont have to march one step at a time anymore. Training lights up the whole sequence at once. Throughput goes way up iteration cycles shrink. - Multi-head attention = multiple viewpoints. The model learns to focus on different relationships simultaneously. Syntax semantics"  
[X Link](https://x.com/willccbb/status/1968151952623272029)  2025-09-17T03:16Z 39.6K followers, 324.4K engagements


"is it just me or is this deepseek paper reallyweird like the flagship results are all about compression ratios and theyre gesturing at implications for LLM memory but its an OCR model are they suggesting that LLMs should ingest OCR embeddings of screenshots of old notes"  
[X Link](https://x.com/willccbb/status/1980160732236042604)  2025-10-20T06:34Z 39.5K followers, 115.4K engagements


"TLDR: codex is so good that people kept trying to use it for harder tasks and it didnt do those as well on those and then people just assumed the model got worse We promised unprecedented transparency for Codex and to take the reports of degradation seriously despite seeing incredible growth week over week. Here is our report and what we have found over the last seven days https://t.co/lJTCAeFHVQ We promised unprecedented transparency for Codex and to take the reports of degradation seriously despite seeing incredible growth week over week. Here is our report and what we have found over the"  
[X Link](https://x.com/willccbb/status/1984884814185746892)  2025-11-02T07:26Z 39.6K followers, 251.3K engagements


"FYI its totally possible to go from publicizing a new OSS project to getting an offer from a lab in like a week"  
[X Link](https://x.com/willccbb/status/2014404393215688902)  2026-01-22T18:26Z 39.6K followers, 80.6K engagements


"consider @primeintellect compute for your next frontier model run. smoothly operating brand new hardware at scale over the holidays isnt for the faint of heart. were here to help And lastly no comments needed here doesnt matter the scale PI got it covered "We trained it on [----] B300s for a month on @PrimeIntellect infrastructure." And lastly no comments needed here doesnt matter the scale PI got it covered "We trained it on [----] B300s for a month on @PrimeIntellect infrastructure.""  
[X Link](https://x.com/willccbb/status/2016391165340463175)  2026-01-28T06:01Z 39.6K followers, 14.9K engagements


"OpenClaw is now MacMiniBot. Due to a Cease and Desist from Apple MacMiniBot is now Moltmax. Due to sounding like a medicine for moths Moltmax is now RedLobster. Due to PE restructuring RedLobster and Red Lobster have merged and your subscription now includes cheesy biscuits"  
[X Link](https://x.com/willccbb/status/2017309439142039699)  2026-01-30T18:50Z 39.6K followers, 175.6K engagements


"california should do tax credits for converting alfalfa farms into datacenters in order to reduce water usage"  
[X Link](https://x.com/willccbb/status/2018037723966398655)  2026-02-01T19:04Z 39.4K followers, 45.8K engagements


"what would it look like to have better incentive structures for open character-training research open models got really good at math and coding but personality is an afterthought. openai and anthropic (correctly) spend real effort on making models that are nice to talk to we dont really have evals for this its hard to measure and thus hard to hillclimb or systematically compare methods. a lot of the work done for closed models is presumably human-data-intensive which is tricky to source in the open. we had all the alignment/RLHF stuff from 2023-24 which has mostly been neglected since RLVR"  
[X Link](https://x.com/willccbb/status/2018060932723995087)  2026-02-01T20:36Z 39.4K followers, 25.3K engagements


"excellent writeup digging into RL training with our hosted platform :) my first blog is up :) https://t.co/6QLoqT2LLd my first blog is up :) https://t.co/6QLoqT2LLd"  
[X Link](https://x.com/willccbb/status/2018195649410580932)  2026-02-02T05:31Z 39.6K followers, 33.6K engagements


"the infra that enables you to A/B test models or prompts is basically the same infra that lets you do reinforcement learning"  
[X Link](https://x.com/willccbb/status/2018746278637871594)  2026-02-03T17:59Z 39.6K followers, 15.3K engagements


"claude code is insane bc it's like codex but for one-off side projects"  
[X Link](https://x.com/willccbb/status/2018750576721576230)  2026-02-03T18:16Z 39.6K followers, 42K engagements


"its the best time ever to be a junior researcher with high ambition and slope regardless of whats on your resume we dont really care about your publications or degrees or past employers just show us something that demonstrates taste agency and ability to execute"  
[X Link](https://x.com/willccbb/status/2018781596154028138)  2026-02-03T20:20Z 39.6K followers, 48.3K engagements


"weve interviewed a lot of people with ex big lab experience or fancy academic pedigrees but havent hired too many people fitting that profile. usually not a culture fit. working at prime intellect is the often most impressive thing on our resumes"  
[X Link](https://x.com/willccbb/status/2018782525687230667)  2026-02-03T20:23Z 39.6K followers, [----] engagements


"me: hey can you do X codex: here's how someone could hypothetically do X Y and Z. if you want me to do Z just say the word"  
[X Link](https://x.com/willccbb/status/2018879467691937901)  2026-02-04T02:49Z 39.6K followers, 14.3K engagements


"monday was a much more important day for agentic coding releases than today"  
[X Link](https://x.com/willccbb/status/2019678827845415196)  2026-02-06T07:45Z 39.4K followers, 32.5K engagements


"the codex desktop app is so good that i literally don't care about whether or not [---] opus is a better model"  
[X Link](https://x.com/willccbb/status/2019685651793514579)  2026-02-06T08:12Z 39.6K followers, 21.9K engagements


"i tried firing off a cloud task from the claude code app and it told me to install github desktop. excuse me"  
[X Link](https://x.com/willccbb/status/2019686429492924448)  2026-02-06T08:15Z 39.4K followers, [----] engagements


"seems like you can extend this to composite rewards (fraction of test cases judge criteria etc) by viewing as a mixture of binary tasks each of which has a policy gradient then just compute the truncated pass@k series for each and take the mixture of MaxRL gradients I haven't been as excited about a paper in a long time but MaxRL (terrible name btw) is really elegant. The jury is still out on how this scales but I will definitely do some ablations with this. https://t.co/eE7cLkwpDP I haven't been as excited about a paper in a long time but MaxRL (terrible name btw) is really elegant. The jury"  
[X Link](https://x.com/willccbb/status/2019788051191366026)  2026-02-06T14:59Z 39.6K followers, 20.3K engagements


"hey man that's quite the prime lab setup you got there"  
[X Link](https://x.com/willccbb/status/2019789468132331854)  2026-02-06T15:05Z 39.4K followers, [----] engagements


"last night was a claude code awakening"  
[X Link](https://x.com/willccbb/status/2019879882805899543)  2026-02-06T21:04Z 39.4K followers, [----] engagements


"not claude code itself. just the concept of claude code. dozens of claudes running around my computer doing useful work. all powered by gpt-5.3-codex-xhigh"  
[X Link](https://x.com/willccbb/status/2019880061890097298)  2026-02-06T21:05Z 39.4K followers, [----] engagements


"road to v0.1.10"  
[X Link](https://x.com/willccbb/status/2019880523779436559)  2026-02-06T21:06Z 39.4K followers, [----] engagements


"i'm going all in on doordash investing $30 in lunch BREAKING: France has gone all-in on science investing [--] million in AI health and climate initiatives. BREAKING: France has gone all-in on science investing [--] million in AI health and climate initiatives"  
[X Link](https://x.com/willccbb/status/2019975035533037592)  2026-02-07T03:22Z 39.6K followers, 60.3K engagements


"a lot of products are increasingly gonna look like features"  
[X Link](https://x.com/willccbb/status/2020102172164477262)  2026-02-07T11:47Z 39.4K followers, [----] engagements


"we have RLMs at home"  
[X Link](https://x.com/willccbb/status/2020119009396457576)  2026-02-07T12:54Z 39.6K followers, 11.9K engagements


"mostly been using claude for chatgpt stuff and codex for claude code stuff"  
[X Link](https://x.com/willccbb/status/2020121100227952814)  2026-02-07T13:02Z 39.4K followers, 10.7K engagements


"havent updated my personal site in ages its still got a sonnet [---] ui"  
[X Link](https://x.com/willccbb/status/2020149263649366040)  2026-02-07T14:54Z 39.6K followers, [----] engagements


"@shiraeis lower batch size"  
[X Link](https://x.com/willccbb/status/2020247435998822553)  2026-02-07T21:24Z 39.4K followers, [----] engagements


"finally figured out the answer to the age-old question of "are tools part of the agent or part of the environment" the answer is "yes""  
[X Link](https://x.com/willccbb/status/2020301302027415995)  2026-02-08T00:59Z 39.5K followers, 15K engagements


"@HeerProfessor problem is that the contract between model and harness is mediated through tool calling"  
[X Link](https://x.com/willccbb/status/2020306510660923687)  2026-02-08T01:19Z 39.4K followers, [---] engagements


"it's abstractions all the way down. abstractions for how to resolve two URLs that should be treated as logically equivalent from the perspective of a program when joined with URL-specific metadata. abstractions for swarm superintelligence. we're hiring"  
[X Link](https://x.com/willccbb/status/2020313424786489500)  2026-02-08T01:47Z 39.4K followers, [----] engagements


"@hallerite harnesses can have prompts skills and tool descriptions which can be part of a trainable policy :)"  
[X Link](https://x.com/willccbb/status/2020317490338058486)  2026-02-08T02:03Z 39.4K followers, [----] engagements


"@hallerite so can an agency like gas town"  
[X Link](https://x.com/willccbb/status/2020317677315887384)  2026-02-08T02:04Z 39.4K followers, [----] engagements


"@HanchungLee is the harness part of the environment in many cases no"  
[X Link](https://x.com/willccbb/status/2020318805218718105)  2026-02-08T02:08Z 39.4K followers, [---] engagements


"@xeophon @rasdani_ did you know we support gepa optimization and also openenv and also byo cli agent"  
[X Link](https://x.com/willccbb/status/2020446754030625277)  2026-02-08T10:36Z 39.4K followers, [---] engagements


"@xeophon @rasdani_ we also have native support for finetuning amazon nova in aws sagemaker"  
[X Link](https://x.com/willccbb/status/2020446936252186834)  2026-02-08T10:37Z 39.4K followers, [---] engagements


"@xeophon @rasdani_ and tinker/skyrl/rllm + (coming soon) nemo & slime"  
[X Link](https://x.com/willccbb/status/2020447251282178149)  2026-02-08T10:38Z 39.4K followers, [---] engagements


"bruh @stalkermustang @PrimeIntellect my RL Envs more than half of them have been pushed in the last [--] days. https://t.co/Hu54Rb20NA @stalkermustang @PrimeIntellect my RL Envs more than half of them have been pushed in the last [--] days. https://t.co/Hu54Rb20NA"  
[X Link](https://x.com/willccbb/status/2020488561036816795)  2026-02-08T13:23Z 39.5K followers, 12.3K engagements


"@iScienceLuvr i *just* started using worktrees i dont know what switch does i dont fully understand origin vs upstream"  
[X Link](https://x.com/willccbb/status/2020507867007295694)  2026-02-08T14:39Z 39.6K followers, [----] engagements


"@menhguin think we can get anthropic client support over the finish line today"  
[X Link](https://x.com/willccbb/status/2020611866121372072)  2026-02-08T21:33Z 39.4K followers, [---] engagements


"@kimbochen imo its not so much inference is too expensive but that training is cheap enough that recomputation is fine and trying to take it from 25%-10% of theoretical cost (ignoring hardware/shapes) isnt worth the effort especially for smaller orgs"  
[X Link](https://x.com/willccbb/status/2020630454634995999)  2026-02-08T22:46Z 39.6K followers, [----] engagements


"featuring models from: - @huggingface - @alibaba_qwen - @allen_ai - @arcee_ai - @meta - @nvidia - @zai_org - @primeintellect initial multimodal support for qwen. more models coming soon. training and inference with adapters on shared infrastructure. Lab unifies everything you need for post-training research into one platform + Environments Hub + Hosted Evaluations + Hosted Training + Deployments & Inference Without needing to worry about the costs of massive GPU clusters or the headaches of low-level algorithm details https://t.co/jYy5UmzQgS Lab unifies everything you need for post-training"  
[X Link](https://x.com/willccbb/status/2021417530456518931)  2026-02-11T02:54Z 39.6K followers, 20.3K engagements


"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect it's still free for the time being this is now "Public Beta" -- we have auto-scaling in place for target load levels but you sometimes get "bonus" speedups if you're doing a run when load is lighter. we'll be watching usage/signups + will announce pricing in advance"  
[X Link](https://x.com/willccbb/status/2021420199405351044)  2026-02-11T03:05Z 39.6K followers, [---] engagements


"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect for more granular control over batch size / resource allocation best bet is to self-deploy prime-rl onto an on-demand or reserved cluster"  
[X Link](https://x.com/willccbb/status/2021420996855726113)  2026-02-11T03:08Z 39.6K followers, [--] engagements


"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect though if a run started fast and has slowed it's also possible that the model is just giving longer responses and thus taking more time per step -- can check on this if you'd like"  
[X Link](https://x.com/willccbb/status/2021421212002550083)  2026-02-11T03:09Z 39.6K followers, [--] engagements


"@osoleve @shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect when were out of beta itll be per-token pricing for training + inference"  
[X Link](https://x.com/willccbb/status/2021426584058826824)  2026-02-11T03:30Z 39.6K followers, [--] engagements


"@0xBebis_ it's a community platform along the lines of github/huggingface moreso than a marketplace we do basic CI tests for builds and you're encouraged to run your own evals + inspect implementations before relying on them this is a basic env template. it's pretty plug and play"  
[X Link](https://x.com/willccbb/status/2021432657062891918)  2026-02-11T03:54Z 39.6K followers, [---] engagements


"are you feeling it @FalduFenil12 @np_hard @Nitish_singla_ cited @willccbb @manveerxyz already loving this. the brainstorm skill gave me a clear path to translate my local work onto prime. really slick https://t.co/xmSOnuIWqG @willccbb @manveerxyz already loving this. the brainstorm skill gave me a clear path to translate my local work onto prime. really slick https://t.co/xmSOnuIWqG"  
[X Link](https://x.com/willccbb/status/2021450732122304781)  2026-02-11T05:06Z 39.6K followers, [----] engagements


"@xeophon lmk what spec u need + well work to make sure its well-supported 🙏"  
[X Link](https://x.com/willccbb/status/2021452108860621004)  2026-02-11T05:11Z 39.6K followers, [---] engagements


"@navneet_rabdiya @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect were routing fwd + bwd no merging. concurrency goes up/down all the time depending on model + load"  
[X Link](https://x.com/willccbb/status/2021458280099573894)  2026-02-11T05:36Z 39.6K followers, [---] engagements


"@xlr8harder @PrimeIntellect yes we have a sandbox layer we use for this (team is running a lot of experiments around opencode + harbor) and you can also bring your own within the rollout loop (e.g many envs use modal for quick GPU execution)"  
[X Link](https://x.com/willccbb/status/2021468174399742463)  2026-02-11T06:15Z 39.6K followers, [---] engagements


"if you want to learn how we built the multi-tenant LoRA orchestration for training and inference with large MoEs just tell your agent to clone prime-rl and explain it to you open superintelligence stack Infrastructure Lab is built on the same stack we used to train INTELLECT-3. Each run gets a dedicated orchestrator with multi-tenant LoRA for training and inference. Enabling shared hardware across runs high efficiency per-token pricing. https://t.co/XpTiWCdd1X Infrastructure Lab is built on the same stack we used to train INTELLECT-3. Each run gets a dedicated orchestrator with multi-tenant"  
[X Link](https://x.com/willccbb/status/2021472924545450486)  2026-02-11T06:34Z 39.6K followers, 19.5K engagements


"@teilomillet @jessicafeiyali just web app no native apps yet but maybe one day"  
[X Link](https://x.com/willccbb/status/2021646728605446243)  2026-02-11T18:05Z 39.6K followers, [--] engagements


"the level of ambition for what we are building towards is quite high. if i told you you'd think i'm going a little crazy. crazy is what got us here. the whole world is crazy. it's gonna be fast and weird and intense. but it's doable. let's chat about the future we can build"  
[X Link](https://x.com/willccbb/status/2021712115216920774)  2026-02-11T22:25Z 39.6K followers, [----] engagements


"none of this existed a year ago. i was fresh out of grad school working at a bank tinkering with the first version of verifiers imagining a future where training your own agents was easier. the prime intellect team was building the first version of prime-rl and proving that you could scale async RL across data centers all around the world. now we have a product and you can use it to do the same. it's easy. your agent can answer all your questions about how to use it. it's open-source. there will be many more products aimed at bringing research to the masses giving all of us more control and"  
[X Link](https://x.com/willccbb/status/2021715583465869449)  2026-02-11T22:38Z 39.6K followers, [----] engagements


"@garybasin spacex"  
[X Link](https://x.com/willccbb/status/2021802008618185090)  2026-02-12T04:22Z 39.6K followers, [----] engagements


"@WindOfNatureAI @PrimeIntellect for now yes this is public beta limited capacity / first come first serve"  
[X Link](https://x.com/willccbb/status/2021813739096617138)  2026-02-12T05:08Z 39.6K followers, [--] engagements


"why are they all moving to private equity. is banking dead"  
[X Link](https://x.com/willccbb/status/2021817506806346125)  2026-02-12T05:23Z 39.6K followers, [----] engagements


"@shiels_ai @PrimeIntellect big fan of the Mutual Information vids on YouTube as a 3b1b style speedrun https://youtube.com/playlistlist=PLzvYlJMoZ02Dxtwe-MmH4nOB5jYlMGBjr&si=hZsu1RyCl1oRZL1c https://youtube.com/playlistlist=PLzvYlJMoZ02Dxtwe-MmH4nOB5jYlMGBjr&si=hZsu1RyCl1oRZL1c"  
[X Link](https://x.com/willccbb/status/2021877083514056740)  2026-02-12T09:20Z 39.6K followers, [----] engagements


"@MaziyarPanahi @prithivMLmods plan is to do one of the big ones (k2.5 glm-5 maybe dsv4 if it drops) in the near future when we ramp capacity a bit more + go GA (still free public beta rn) wont be free haha"  
[X Link](https://x.com/willccbb/status/2022027248220745861)  2026-02-12T19:17Z 39.6K followers, [--] engagements


"@DagmawiBabi training is currently free in the public beta phase well share pricing info shortly with a heads-up before we enable billing. the free period allows us to see utilization patterns as we finalize numbers"  
[X Link](https://x.com/willccbb/status/2022033450627465277)  2026-02-12T19:41Z 39.6K followers, [---] engagements


"it's interesting that [---] billion parameters is all you need to crush math competitions but you need like [--] trillion to make the model be funny maybe humor is the right measure of true intelligence"  
[X Link](https://x.com/willccbb/status/1895504679674773626)  2025-02-28T16:01Z 39.6K followers, 12.8M engagements


"@giffmana gemini is a VLA trained on lots of video so it's basically JEPA. huge yann lecun cultural victory"  
[X Link](https://x.com/willccbb/status/2019169522419397062)  2026-02-04T22:01Z 39.6K followers, [----] engagements


"i really shouldn't have to tell the "best coding model in the world" to write better code so often. it's in a repo full of approved code. it has plenty of context on what i think good code is. and yet it still writes bad code constantly monday was a much more important day for agentic coding releases than today monday was a much more important day for agentic coding releases than today"  
[X Link](https://x.com/willccbb/status/2019682529549381888)  2026-02-06T08:00Z 39.6K followers, 26.7K engagements


"average Anthropic ad: MIDNIGHT https://t.co/XrXx5OwqDZ MIDNIGHT https://t.co/XrXx5OwqDZ"  
[X Link](https://x.com/willccbb/status/2019878427046866985)  2026-02-06T20:58Z 39.6K followers, 15.5K engagements


"(seeing a bunch of analysts quit goldman sachs in february) omg what's happening at goldman"  
[X Link](https://x.com/willccbb/status/2021817305773334866)  2026-02-12T05:23Z 39.6K followers, [----] engagements


"sometimes thats what you want. other times it very much isnt"  
[X Link](https://x.com/willccbb/status/2022011810615026020)  2026-02-12T18:15Z 39.6K followers, [----] engagements


"i asked claude if he would do this and he said probably not JUST IN - Pentagon used Anthropics AI tool Claude in the military operation to capture Nicolas Maduro WSJ https://t.co/BcHtsInSDU JUST IN - Pentagon used Anthropics AI tool Claude in the military operation to capture Nicolas Maduro WSJ https://t.co/BcHtsInSDU"  
[X Link](https://x.com/willccbb/status/2022457900510384415)  2026-02-13T23:48Z 39.6K followers, 22.6K engagements


"we need inverse MCP handoff protocol where the model makes the user do tool calls"  
[X Link](https://x.com/willccbb/status/2020079619677065496)  2026-02-07T10:18Z 39.6K followers, [----] engagements


"create your own environments. train your own models. be your own lab. Introducing Lab: A full-stack platform for training your own agentic models Build evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab. https://t.co/wDVCe7TOdt Introducing Lab: A full-stack platform for training your own agentic models Build evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab. https://t.co/wDVCe7TOdt"  
[X Link](https://x.com/willccbb/status/2021410167523508368)  2026-02-11T02:25Z 39.6K followers, 97.9K engagements


"the most important thing for you to be doing right now is having good ideas"  
[X Link](https://x.com/willccbb/status/2021652583069495698)  2026-02-11T18:28Z 39.6K followers, 39.3K engagements


"with the latest models i am now finding myself thinking about complex system design problems more not less. the magnitude of what can be reasonably attempted is monumentally larger. you need to make sure youre asking it to build the right thing"  
[X Link](https://x.com/willccbb/status/2021653634828972447)  2026-02-11T18:32Z 39.6K followers, 35.9K engagements


"opus is like if codex could make eye contact"  
[X Link](https://x.com/willccbb/status/2022010717801787877)  2026-02-12T18:11Z 39.6K followers, 128.3K engagements


"Deployments are live train a model then immediately use it for inference"  
[X Link](https://x.com/willccbb/status/2022030806798189029)  2026-02-12T19:31Z 39.6K followers, [----] engagements


"380B (30B active) Anthropic raised $30B and is now valued at $380B https://t.co/An0AG4HX39 Anthropic raised $30B and is now valued at $380B https://t.co/An0AG4HX39"  
[X Link](https://x.com/willccbb/status/2022036359385690151)  2026-02-12T19:53Z 39.6K followers, 92.5K engagements


"@medlockmillions but what if the superduper ai invents a new kind of infinite money we cant even fathom with our little ant brains"  
[X Link](https://x.com/willccbb/status/2022492202371584343)  2026-02-14T02:04Z 39.6K followers, [----] engagements


"turing machine is just a FSM with a tape value is much more in the conceptual framing than the actual capabilities delta with current agents some work is serial some is parallelizable in theory you want a "general agent" to be able to process depth-first vs breadth-first at will delegating requisite context down the tree (this has maybe been semi-obvious to people who spent a lot of time thinking about agents but it didn't have a name or much explicit focus beyond "agent with subagents and maybe the subagents have other subagents ad infinitum")"  
[X Link](https://x.com/willccbb/status/2022497671882420408)  2026-02-14T02:26Z 39.6K followers, [----] engagements


"@teilomillet its really the model thats getting the gradients not the tokens. ends up being practical to have a model-level objective which can be linearly decomposed at a token level but the interpretation isnt as shallow as just good token/bad token"  
[X Link](https://x.com/willccbb/status/2022875086030803452)  2026-02-15T03:26Z 39.6K followers, [----] engagements


"pro tip for codex desktop: set up shortcuts for real terminal + real text editor"  
[X Link](https://x.com/willccbb/status/2022877409100648769)  2026-02-15T03:35Z 39.6K followers, 10.2K engagements


"@adonis_singh sonnet 3.6"  
[X Link](https://x.com/willccbb/status/2023106672278090094)  2026-02-15T18:46Z 39.6K followers, [---] engagements


"@turbo_xo_ this is also a nice interface within the diff viewer for the types of small edits where a text editor is preferable over full agentic chat"  
[X Link](https://x.com/willccbb/status/2023129136982852033)  2026-02-15T20:15Z 39.6K followers, [--] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing

@willccbb Avatar @willccbb will brown

will brown posts on X about in the, ai, inference, model the most. They currently have [------] followers and [---] posts still getting attention that total [------] engagements in the last [--] hours.

Engagements: [------] #

Engagements Line Chart

  • [--] Week [-------] -31%
  • [--] Month [---------] +82%
  • [--] Months [----------] -45%
  • [--] Year [----------] +797%

Mentions: [--] #

Mentions Line Chart

  • [--] Week [---] +6.90%
  • [--] Month [---] +100%
  • [--] Months [---] +24%
  • [--] Year [-----] +385%

Followers: [------] #

Followers Line Chart

  • [--] Week [------] +0.77%
  • [--] Month [------] +3.50%
  • [--] Months [------] +32%
  • [--] Year [------] +234%

CreatorRank: [-------] #

CreatorRank Line Chart

Social Influence

Social category influence technology brands 22.52% stocks 9.91% finance 5.41% social networks 1.8% products 0.9% countries 0.9% financial services 0.9%

Social topic influence in the 8.11%, ai 7.21%, inference #235, model #1473, meta 5.41%, agentic #494, anthropic #1734, future 3.6%, lab #1018, claude code 3.6%

Top accounts mentioned or mentioned by @primeintellect @xeophon @maziyarpanahi @huggingface @alibabaqwen @allenai @arceeai @meta @nvidia @zaiorg @shielsai @shiels_ai @hallerite @osoleve @faldufenil12 @teilomillet @heerprofessor @rasdani @hanchunglee @stalkermustang

Top assets mentioned Alphabet Inc Class A (GOOGL) Morgan Stanley (MS) DoorDash Inc. (DASH)

Top Social Posts

Top posts by engagements in the last [--] hours

"your research artifacts need to be Legible to Clankers"
X Link 2026-02-11T19:29Z 39.6K followers, [----] engagements

"the future belongs to those who vaguely understand git"
X Link 2026-02-07T11:39Z 39.6K followers, 272.5K engagements

"how it feels to be deep in the thralls of gpt-5.3-codex psychosis"
X Link 2026-02-08T00:16Z 39.6K followers, 10.1K engagements

"consider ask your employer if your token budget includes training tokens as well as inference tokens"
X Link 2026-02-11T22:09Z 39.6K followers, [----] engagements

"dwarkesh: you should buy more compute dario: were buying a lot dwarkesh: what if its not enough dario: then we make profit. if we buy too much we go bankrupt dwarkesh: but what about the country of geniuses in a datacenter dario: theyll have a nice big datacenter"
X Link 2026-02-13T20:55Z 39.6K followers, 150K engagements

"imo doesnt really make sense to think of it as operating literally at just the token level. the objective is to maximize the relative likelihoods of the rollouts with higher rewards. tokens are just one intermediate representation. mechanistically this looks like identifying the circuit chains whose activations are most correlated with observed rewards. even in the case where all rollouts branch in the first few tokens you have higher-level properties (strategies subsequences) which are more present in the better rollouts and these are captured by latent + sequence-level representations."
X Link 2026-02-15T03:23Z 39.6K followers, [----] engagements

"if you're a later-year PhD student thinking about internships for next summer my team at Morgan Stanley (ML Research) is hiring we're a tight-knit team of [--] researchers working on everything from time-series to RL to diffusion to LLMs both applications and fundamentals. we have compute publishing freedom and an endless supply of hard and interesting problems. feel free to reach out to me with any questions"
X Link 2024-08-18T20:00Z 39.6K followers, 337.8K engagements

"if youre doing a PhD in a quantitative discipline you should print this out and hang it above your desk https://www.lkozma.net/inequalities_cheat_sheet/ineq.pdf https://www.lkozma.net/inequalities_cheat_sheet/ineq.pdf"
X Link 2024-09-05T12:54Z 39.6K followers, 161.2K engagements

"i first got into agentic programming about [--] years ago"
X Link 2024-12-16T03:37Z 39.6K followers, 57.1K engagements

"LLMs know way too many facts. you can ask Llama 1B about the history of pizza and it gives a decent answer. it shouldn't be able to do that. it should just do a google search. use those weights for something else"
X Link 2025-02-06T13:10Z 39.6K followers, 307K engagements

"Of course thats your contention. Youre a first-year AI engineer who learned the hard way that RAG is more than vector search. You just finished reading Building effective agents and were convinced to migrate from LangChain to Model Complex Protocol until next month when"
X Link 2025-03-06T20:35Z 39.6K followers, 140.7K engagements

"coding on your iphone: - acquire mac mini - set up dotfiles nvim etc - install ssh client on phone - ssh to mini - npm install -g @anthropic-ai/claude-code - scripts/tools for web search e2b/docker cli chat - use git branches + aggressively - profit http://CLAUDE.md http://CLAUDE.md"
X Link 2025-03-09T03:05Z 39.6K followers, 181.7K engagements

"11k youtube views on this already is kinda wild to me lol thanks for checkin it out :)"
X Link 2025-03-09T21:22Z 39.6K followers, 133.8K engagements

"the new llama [--] models are so advanced that they require versions of hf transformers that haven't even been invented yet"
X Link 2025-04-05T19:42Z 39.6K followers, 164.9K engagements

"update: i joined @primeintellect :) cannot describe how excited i am to be joining such an incredible team and mission. there is a dire shortage of labs who are truly embracing open-source research. its hard to get the incentives right. you need a business model where open-sourcing your work is positive-sum; being a GPU marketplace is a really good one. prime intellect has been doing incredible work to advance the frontiers of decentralized training and inference and it is only the beginning. my own goal is to continue along the directions of my recent projects and musings but bigger bolder"
X Link 2025-05-08T17:56Z 39.6K followers, 214.6K engagements

"the year is [----]. python [---] now compiles directly to Rython borrow-checked zero-cost no more gc. pip defaults to uv add and tokio ships with stdlib. the GIL has been replaced with lifetimes and del wraps std::mem::forget. we still argue about tabs vs spaces. Today were announcing the preview release of ty an extremely fast type checker and language server for Python written in Rust. In early testing it's 10x 50x even 100x faster than existing type checkers. (We've seen 600x speed-ups over Mypy in some real-world projects.) https://t.co/bAiygYPhqP Today were announcing the preview release of"
X Link 2025-05-13T17:16Z 39.6K followers, 92.8K engagements

"happy saturday for the price of a cup of coffee you can rent a [----] for the day and do some random experiments on small models"
X Link 2025-05-17T17:29Z 39.6K followers, 294K engagements

"being good at ML systems helps you run more experiments. being good at ML theory helps you run less experiments"
X Link 2025-06-02T02:29Z 39.6K followers, 94.9K engagements

"you can just RL a model to play Wordle"
X Link 2025-06-03T18:58Z 39.6K followers, 131.3K engagements

"you can literally make [--] figures for just having a hobby hobbies are just excuses for unambitious people to fill their free time hobbies are just excuses for unambitious people to fill their free time"
X Link 2025-06-04T00:48Z 39.6K followers, 429K engagements

"spending everything on Alexandr Wang"
X Link 2025-06-10T15:45Z 39.6K followers, 120.1K engagements

"2+ years of MCP experience required"
X Link 2025-06-25T19:27Z 39.6K followers, 70.8K engagements

"xAI Head of Product announces that Grok [--] is the Antichrist Going from an office where AI researchers are building the Antichrist to my living room where my girlfriend is watching Love Island is one of the most drastic transitions in the known universe Going from an office where AI researchers are building the Antichrist to my living room where my girlfriend is watching Love Island is one of the most drastic transitions in the known universe"
X Link 2025-07-08T09:10Z 39.6K followers, 204.9K engagements

"and were live been a very long time in the making huge thanks to everyone whos made it possible along the way. cant wait to see what you guys all build here. were just getting started :) Introducing the Environments Hub RL environments are the key bottleneck to the next wave of AI progress but big labs are locking them down We built a community platform for crowdsourcing open environments so anyone can contribute to open-source AGI https://t.co/urAv2hGPCQ Introducing the Environments Hub RL environments are the key bottleneck to the next wave of AI progress but big labs are locking them down"
X Link 2025-08-27T19:21Z 39.5K followers, 295.2K engagements

"were hiring for lots of junior roles right now. apprenticeships basically. can be part-time remote international. if you really want to get our attention for these share a cool env to the hub [---] LOC. well give full-time offers to top performers after. we can sponsor. Called it last year. Junior roles down 23%. Senior roles up 14%. Harvard study tracked [------] firms. Results: Before AI: [--] senior + [--] juniors = [--] person team After AI: [--] senior + Claude = same output https://t.co/wAdQz9Sg8Z Called it last year. Junior roles down 23%. Senior roles up 14%. Harvard study tracked [------] firms."
X Link 2025-09-06T13:58Z 39.6K followers, 419K engagements

"now this is a hiring page Excited to share that I am starting a new company dedicated to the creation of verified superintelligence via autoformalization. Building on the amazing RL infrastructure that we have developed at @morph_labs @mathematics_inc has already achieved a breakthrough result by Excited to share that I am starting a new company dedicated to the creation of verified superintelligence via autoformalization. Building on the amazing RL infrastructure that we have developed at @morph_labs @mathematics_inc has already achieved a breakthrough result by"
X Link 2025-09-11T18:18Z 39.6K followers, 357.9K engagements

"I just read this new paper from Google and Im absolutely buzzing 🤯 The core idea is almost offensively simple: ditch recurrence and convolutions and use only attention. Thats it. And somehowit unlocks a whole new regime of performance scale and simplicity. Heres what blew my mind: - No recurrence full parallelism. Tokens dont have to march one step at a time anymore. Training lights up the whole sequence at once. Throughput goes way up iteration cycles shrink. - Multi-head attention = multiple viewpoints. The model learns to focus on different relationships simultaneously. Syntax semantics"
X Link 2025-09-17T03:16Z 39.6K followers, 324.4K engagements

"is it just me or is this deepseek paper reallyweird like the flagship results are all about compression ratios and theyre gesturing at implications for LLM memory but its an OCR model are they suggesting that LLMs should ingest OCR embeddings of screenshots of old notes"
X Link 2025-10-20T06:34Z 39.5K followers, 115.4K engagements

"TLDR: codex is so good that people kept trying to use it for harder tasks and it didnt do those as well on those and then people just assumed the model got worse We promised unprecedented transparency for Codex and to take the reports of degradation seriously despite seeing incredible growth week over week. Here is our report and what we have found over the last seven days https://t.co/lJTCAeFHVQ We promised unprecedented transparency for Codex and to take the reports of degradation seriously despite seeing incredible growth week over week. Here is our report and what we have found over the"
X Link 2025-11-02T07:26Z 39.6K followers, 251.3K engagements

"FYI its totally possible to go from publicizing a new OSS project to getting an offer from a lab in like a week"
X Link 2026-01-22T18:26Z 39.6K followers, 80.6K engagements

"consider @primeintellect compute for your next frontier model run. smoothly operating brand new hardware at scale over the holidays isnt for the faint of heart. were here to help And lastly no comments needed here doesnt matter the scale PI got it covered "We trained it on [----] B300s for a month on @PrimeIntellect infrastructure." And lastly no comments needed here doesnt matter the scale PI got it covered "We trained it on [----] B300s for a month on @PrimeIntellect infrastructure.""
X Link 2026-01-28T06:01Z 39.6K followers, 14.9K engagements

"OpenClaw is now MacMiniBot. Due to a Cease and Desist from Apple MacMiniBot is now Moltmax. Due to sounding like a medicine for moths Moltmax is now RedLobster. Due to PE restructuring RedLobster and Red Lobster have merged and your subscription now includes cheesy biscuits"
X Link 2026-01-30T18:50Z 39.6K followers, 175.6K engagements

"california should do tax credits for converting alfalfa farms into datacenters in order to reduce water usage"
X Link 2026-02-01T19:04Z 39.4K followers, 45.8K engagements

"what would it look like to have better incentive structures for open character-training research open models got really good at math and coding but personality is an afterthought. openai and anthropic (correctly) spend real effort on making models that are nice to talk to we dont really have evals for this its hard to measure and thus hard to hillclimb or systematically compare methods. a lot of the work done for closed models is presumably human-data-intensive which is tricky to source in the open. we had all the alignment/RLHF stuff from 2023-24 which has mostly been neglected since RLVR"
X Link 2026-02-01T20:36Z 39.4K followers, 25.3K engagements

"excellent writeup digging into RL training with our hosted platform :) my first blog is up :) https://t.co/6QLoqT2LLd my first blog is up :) https://t.co/6QLoqT2LLd"
X Link 2026-02-02T05:31Z 39.6K followers, 33.6K engagements

"the infra that enables you to A/B test models or prompts is basically the same infra that lets you do reinforcement learning"
X Link 2026-02-03T17:59Z 39.6K followers, 15.3K engagements

"claude code is insane bc it's like codex but for one-off side projects"
X Link 2026-02-03T18:16Z 39.6K followers, 42K engagements

"its the best time ever to be a junior researcher with high ambition and slope regardless of whats on your resume we dont really care about your publications or degrees or past employers just show us something that demonstrates taste agency and ability to execute"
X Link 2026-02-03T20:20Z 39.6K followers, 48.3K engagements

"weve interviewed a lot of people with ex big lab experience or fancy academic pedigrees but havent hired too many people fitting that profile. usually not a culture fit. working at prime intellect is the often most impressive thing on our resumes"
X Link 2026-02-03T20:23Z 39.6K followers, [----] engagements

"me: hey can you do X codex: here's how someone could hypothetically do X Y and Z. if you want me to do Z just say the word"
X Link 2026-02-04T02:49Z 39.6K followers, 14.3K engagements

"monday was a much more important day for agentic coding releases than today"
X Link 2026-02-06T07:45Z 39.4K followers, 32.5K engagements

"the codex desktop app is so good that i literally don't care about whether or not [---] opus is a better model"
X Link 2026-02-06T08:12Z 39.6K followers, 21.9K engagements

"i tried firing off a cloud task from the claude code app and it told me to install github desktop. excuse me"
X Link 2026-02-06T08:15Z 39.4K followers, [----] engagements

"seems like you can extend this to composite rewards (fraction of test cases judge criteria etc) by viewing as a mixture of binary tasks each of which has a policy gradient then just compute the truncated pass@k series for each and take the mixture of MaxRL gradients I haven't been as excited about a paper in a long time but MaxRL (terrible name btw) is really elegant. The jury is still out on how this scales but I will definitely do some ablations with this. https://t.co/eE7cLkwpDP I haven't been as excited about a paper in a long time but MaxRL (terrible name btw) is really elegant. The jury"
X Link 2026-02-06T14:59Z 39.6K followers, 20.3K engagements

"hey man that's quite the prime lab setup you got there"
X Link 2026-02-06T15:05Z 39.4K followers, [----] engagements

"last night was a claude code awakening"
X Link 2026-02-06T21:04Z 39.4K followers, [----] engagements

"not claude code itself. just the concept of claude code. dozens of claudes running around my computer doing useful work. all powered by gpt-5.3-codex-xhigh"
X Link 2026-02-06T21:05Z 39.4K followers, [----] engagements

"road to v0.1.10"
X Link 2026-02-06T21:06Z 39.4K followers, [----] engagements

"i'm going all in on doordash investing $30 in lunch BREAKING: France has gone all-in on science investing [--] million in AI health and climate initiatives. BREAKING: France has gone all-in on science investing [--] million in AI health and climate initiatives"
X Link 2026-02-07T03:22Z 39.6K followers, 60.3K engagements

"a lot of products are increasingly gonna look like features"
X Link 2026-02-07T11:47Z 39.4K followers, [----] engagements

"we have RLMs at home"
X Link 2026-02-07T12:54Z 39.6K followers, 11.9K engagements

"mostly been using claude for chatgpt stuff and codex for claude code stuff"
X Link 2026-02-07T13:02Z 39.4K followers, 10.7K engagements

"havent updated my personal site in ages its still got a sonnet [---] ui"
X Link 2026-02-07T14:54Z 39.6K followers, [----] engagements

"@shiraeis lower batch size"
X Link 2026-02-07T21:24Z 39.4K followers, [----] engagements

"finally figured out the answer to the age-old question of "are tools part of the agent or part of the environment" the answer is "yes""
X Link 2026-02-08T00:59Z 39.5K followers, 15K engagements

"@HeerProfessor problem is that the contract between model and harness is mediated through tool calling"
X Link 2026-02-08T01:19Z 39.4K followers, [---] engagements

"it's abstractions all the way down. abstractions for how to resolve two URLs that should be treated as logically equivalent from the perspective of a program when joined with URL-specific metadata. abstractions for swarm superintelligence. we're hiring"
X Link 2026-02-08T01:47Z 39.4K followers, [----] engagements

"@hallerite harnesses can have prompts skills and tool descriptions which can be part of a trainable policy :)"
X Link 2026-02-08T02:03Z 39.4K followers, [----] engagements

"@hallerite so can an agency like gas town"
X Link 2026-02-08T02:04Z 39.4K followers, [----] engagements

"@HanchungLee is the harness part of the environment in many cases no"
X Link 2026-02-08T02:08Z 39.4K followers, [---] engagements

"@xeophon @rasdani_ did you know we support gepa optimization and also openenv and also byo cli agent"
X Link 2026-02-08T10:36Z 39.4K followers, [---] engagements

"@xeophon @rasdani_ we also have native support for finetuning amazon nova in aws sagemaker"
X Link 2026-02-08T10:37Z 39.4K followers, [---] engagements

"@xeophon @rasdani_ and tinker/skyrl/rllm + (coming soon) nemo & slime"
X Link 2026-02-08T10:38Z 39.4K followers, [---] engagements

"bruh @stalkermustang @PrimeIntellect my RL Envs more than half of them have been pushed in the last [--] days. https://t.co/Hu54Rb20NA @stalkermustang @PrimeIntellect my RL Envs more than half of them have been pushed in the last [--] days. https://t.co/Hu54Rb20NA"
X Link 2026-02-08T13:23Z 39.5K followers, 12.3K engagements

"@iScienceLuvr i just started using worktrees i dont know what switch does i dont fully understand origin vs upstream"
X Link 2026-02-08T14:39Z 39.6K followers, [----] engagements

"@menhguin think we can get anthropic client support over the finish line today"
X Link 2026-02-08T21:33Z 39.4K followers, [---] engagements

"@kimbochen imo its not so much inference is too expensive but that training is cheap enough that recomputation is fine and trying to take it from 25%-10% of theoretical cost (ignoring hardware/shapes) isnt worth the effort especially for smaller orgs"
X Link 2026-02-08T22:46Z 39.6K followers, [----] engagements

"featuring models from: - @huggingface - @alibaba_qwen - @allen_ai - @arcee_ai - @meta - @nvidia - @zai_org - @primeintellect initial multimodal support for qwen. more models coming soon. training and inference with adapters on shared infrastructure. Lab unifies everything you need for post-training research into one platform + Environments Hub + Hosted Evaluations + Hosted Training + Deployments & Inference Without needing to worry about the costs of massive GPU clusters or the headaches of low-level algorithm details https://t.co/jYy5UmzQgS Lab unifies everything you need for post-training"
X Link 2026-02-11T02:54Z 39.6K followers, 20.3K engagements

"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect it's still free for the time being this is now "Public Beta" -- we have auto-scaling in place for target load levels but you sometimes get "bonus" speedups if you're doing a run when load is lighter. we'll be watching usage/signups + will announce pricing in advance"
X Link 2026-02-11T03:05Z 39.6K followers, [---] engagements

"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect for more granular control over batch size / resource allocation best bet is to self-deploy prime-rl onto an on-demand or reserved cluster"
X Link 2026-02-11T03:08Z 39.6K followers, [--] engagements

"@shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect though if a run started fast and has slowed it's also possible that the model is just giving longer responses and thus taking more time per step -- can check on this if you'd like"
X Link 2026-02-11T03:09Z 39.6K followers, [--] engagements

"@osoleve @shiels_ai @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect when were out of beta itll be per-token pricing for training + inference"
X Link 2026-02-11T03:30Z 39.6K followers, [--] engagements

"@0xBebis_ it's a community platform along the lines of github/huggingface moreso than a marketplace we do basic CI tests for builds and you're encouraged to run your own evals + inspect implementations before relying on them this is a basic env template. it's pretty plug and play"
X Link 2026-02-11T03:54Z 39.6K followers, [---] engagements

"are you feeling it @FalduFenil12 @np_hard @Nitish_singla_ cited @willccbb @manveerxyz already loving this. the brainstorm skill gave me a clear path to translate my local work onto prime. really slick https://t.co/xmSOnuIWqG @willccbb @manveerxyz already loving this. the brainstorm skill gave me a clear path to translate my local work onto prime. really slick https://t.co/xmSOnuIWqG"
X Link 2026-02-11T05:06Z 39.6K followers, [----] engagements

"@xeophon lmk what spec u need + well work to make sure its well-supported 🙏"
X Link 2026-02-11T05:11Z 39.6K followers, [---] engagements

"@navneet_rabdiya @huggingface @Alibaba_Qwen @allen_ai @arcee_ai @Meta @nvidia @Zai_org @PrimeIntellect were routing fwd + bwd no merging. concurrency goes up/down all the time depending on model + load"
X Link 2026-02-11T05:36Z 39.6K followers, [---] engagements

"@xlr8harder @PrimeIntellect yes we have a sandbox layer we use for this (team is running a lot of experiments around opencode + harbor) and you can also bring your own within the rollout loop (e.g many envs use modal for quick GPU execution)"
X Link 2026-02-11T06:15Z 39.6K followers, [---] engagements

"if you want to learn how we built the multi-tenant LoRA orchestration for training and inference with large MoEs just tell your agent to clone prime-rl and explain it to you open superintelligence stack Infrastructure Lab is built on the same stack we used to train INTELLECT-3. Each run gets a dedicated orchestrator with multi-tenant LoRA for training and inference. Enabling shared hardware across runs high efficiency per-token pricing. https://t.co/XpTiWCdd1X Infrastructure Lab is built on the same stack we used to train INTELLECT-3. Each run gets a dedicated orchestrator with multi-tenant"
X Link 2026-02-11T06:34Z 39.6K followers, 19.5K engagements

"@teilomillet @jessicafeiyali just web app no native apps yet but maybe one day"
X Link 2026-02-11T18:05Z 39.6K followers, [--] engagements

"the level of ambition for what we are building towards is quite high. if i told you you'd think i'm going a little crazy. crazy is what got us here. the whole world is crazy. it's gonna be fast and weird and intense. but it's doable. let's chat about the future we can build"
X Link 2026-02-11T22:25Z 39.6K followers, [----] engagements

"none of this existed a year ago. i was fresh out of grad school working at a bank tinkering with the first version of verifiers imagining a future where training your own agents was easier. the prime intellect team was building the first version of prime-rl and proving that you could scale async RL across data centers all around the world. now we have a product and you can use it to do the same. it's easy. your agent can answer all your questions about how to use it. it's open-source. there will be many more products aimed at bringing research to the masses giving all of us more control and"
X Link 2026-02-11T22:38Z 39.6K followers, [----] engagements

"@garybasin spacex"
X Link 2026-02-12T04:22Z 39.6K followers, [----] engagements

"@WindOfNatureAI @PrimeIntellect for now yes this is public beta limited capacity / first come first serve"
X Link 2026-02-12T05:08Z 39.6K followers, [--] engagements

"why are they all moving to private equity. is banking dead"
X Link 2026-02-12T05:23Z 39.6K followers, [----] engagements

"@shiels_ai @PrimeIntellect big fan of the Mutual Information vids on YouTube as a 3b1b style speedrun https://youtube.com/playlistlist=PLzvYlJMoZ02Dxtwe-MmH4nOB5jYlMGBjr&si=hZsu1RyCl1oRZL1c https://youtube.com/playlistlist=PLzvYlJMoZ02Dxtwe-MmH4nOB5jYlMGBjr&si=hZsu1RyCl1oRZL1c"
X Link 2026-02-12T09:20Z 39.6K followers, [----] engagements

"@MaziyarPanahi @prithivMLmods plan is to do one of the big ones (k2.5 glm-5 maybe dsv4 if it drops) in the near future when we ramp capacity a bit more + go GA (still free public beta rn) wont be free haha"
X Link 2026-02-12T19:17Z 39.6K followers, [--] engagements

"@DagmawiBabi training is currently free in the public beta phase well share pricing info shortly with a heads-up before we enable billing. the free period allows us to see utilization patterns as we finalize numbers"
X Link 2026-02-12T19:41Z 39.6K followers, [---] engagements

"it's interesting that [---] billion parameters is all you need to crush math competitions but you need like [--] trillion to make the model be funny maybe humor is the right measure of true intelligence"
X Link 2025-02-28T16:01Z 39.6K followers, 12.8M engagements

"@giffmana gemini is a VLA trained on lots of video so it's basically JEPA. huge yann lecun cultural victory"
X Link 2026-02-04T22:01Z 39.6K followers, [----] engagements

"i really shouldn't have to tell the "best coding model in the world" to write better code so often. it's in a repo full of approved code. it has plenty of context on what i think good code is. and yet it still writes bad code constantly monday was a much more important day for agentic coding releases than today monday was a much more important day for agentic coding releases than today"
X Link 2026-02-06T08:00Z 39.6K followers, 26.7K engagements

"average Anthropic ad: MIDNIGHT https://t.co/XrXx5OwqDZ MIDNIGHT https://t.co/XrXx5OwqDZ"
X Link 2026-02-06T20:58Z 39.6K followers, 15.5K engagements

"(seeing a bunch of analysts quit goldman sachs in february) omg what's happening at goldman"
X Link 2026-02-12T05:23Z 39.6K followers, [----] engagements

"sometimes thats what you want. other times it very much isnt"
X Link 2026-02-12T18:15Z 39.6K followers, [----] engagements

"i asked claude if he would do this and he said probably not JUST IN - Pentagon used Anthropics AI tool Claude in the military operation to capture Nicolas Maduro WSJ https://t.co/BcHtsInSDU JUST IN - Pentagon used Anthropics AI tool Claude in the military operation to capture Nicolas Maduro WSJ https://t.co/BcHtsInSDU"
X Link 2026-02-13T23:48Z 39.6K followers, 22.6K engagements

"we need inverse MCP handoff protocol where the model makes the user do tool calls"
X Link 2026-02-07T10:18Z 39.6K followers, [----] engagements

"create your own environments. train your own models. be your own lab. Introducing Lab: A full-stack platform for training your own agentic models Build evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab. https://t.co/wDVCe7TOdt Introducing Lab: A full-stack platform for training your own agentic models Build evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab. https://t.co/wDVCe7TOdt"
X Link 2026-02-11T02:25Z 39.6K followers, 97.9K engagements

"the most important thing for you to be doing right now is having good ideas"
X Link 2026-02-11T18:28Z 39.6K followers, 39.3K engagements

"with the latest models i am now finding myself thinking about complex system design problems more not less. the magnitude of what can be reasonably attempted is monumentally larger. you need to make sure youre asking it to build the right thing"
X Link 2026-02-11T18:32Z 39.6K followers, 35.9K engagements

"opus is like if codex could make eye contact"
X Link 2026-02-12T18:11Z 39.6K followers, 128.3K engagements

"Deployments are live train a model then immediately use it for inference"
X Link 2026-02-12T19:31Z 39.6K followers, [----] engagements

"380B (30B active) Anthropic raised $30B and is now valued at $380B https://t.co/An0AG4HX39 Anthropic raised $30B and is now valued at $380B https://t.co/An0AG4HX39"
X Link 2026-02-12T19:53Z 39.6K followers, 92.5K engagements

"@medlockmillions but what if the superduper ai invents a new kind of infinite money we cant even fathom with our little ant brains"
X Link 2026-02-14T02:04Z 39.6K followers, [----] engagements

"turing machine is just a FSM with a tape value is much more in the conceptual framing than the actual capabilities delta with current agents some work is serial some is parallelizable in theory you want a "general agent" to be able to process depth-first vs breadth-first at will delegating requisite context down the tree (this has maybe been semi-obvious to people who spent a lot of time thinking about agents but it didn't have a name or much explicit focus beyond "agent with subagents and maybe the subagents have other subagents ad infinitum")"
X Link 2026-02-14T02:26Z 39.6K followers, [----] engagements

"@teilomillet its really the model thats getting the gradients not the tokens. ends up being practical to have a model-level objective which can be linearly decomposed at a token level but the interpretation isnt as shallow as just good token/bad token"
X Link 2026-02-15T03:26Z 39.6K followers, [----] engagements

"pro tip for codex desktop: set up shortcuts for real terminal + real text editor"
X Link 2026-02-15T03:35Z 39.6K followers, 10.2K engagements

"@adonis_singh sonnet 3.6"
X Link 2026-02-15T18:46Z 39.6K followers, [---] engagements

"@turbo_xo_ this is also a nice interface within the diff viewer for the types of small edits where a text editor is preferable over full agentic chat"
X Link 2026-02-15T20:15Z 39.6K followers, [--] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing

@willccbb
/creator/twitter::willccbb