[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

# ![@natolambert Avatar](https://lunarcrush.com/gi/w:26/cr:twitter::2939913921.png) @natolambert Nathan Lambert

Nathan Lambert has been sharing and discussing recent developments in AI, particularly in reinforcement learning (RL) and large language models (LLMs). A new paper on scaling RL with LLMs has been released, and Lambert is positive about its potential, highlighting the importance of techniques like importance sampling and continuous batching. He is also discussing the rise of open models, including those from China, and the need for the US to invest in its own open models.

### Engagements: XXXXXXX [#](/creator/twitter::2939913921/interactions)
![Engagements Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::2939913921/c:line/m:interactions.svg)

- X Week XXXXXXXXX +155%
- X Month XXXXXXXXX +291%
- X Months XXXXXXXXX +16%
- X Year XXXXXXXXXX +199%

### Mentions: XX [#](/creator/twitter::2939913921/posts_active)
![Mentions Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::2939913921/c:line/m:posts_active.svg)

- X Week XX +48%
- X Month XX +30%
- X Months XXX +72%
- X Year XXX +60%

### Followers: XXXXXX [#](/creator/twitter::2939913921/followers)
![Followers Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::2939913921/c:line/m:followers.svg)

- X Week XXXXXX +1.20%
- X Month XXXXXX +3.70%
- X Months XXXXXX +21%
- X Year XXXXXX +101%

### CreatorRank: XXXXXX [#](/creator/twitter::2939913921/influencer_rank)
![CreatorRank Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::2939913921/c:line/m:influencer_rank.svg)

### Social Influence [#](/creator/twitter::2939913921/influence)
---

**Social category influence**
[technology brands](/list/technology-brands)  #2843 [stocks](/list/stocks)  #652 [cryptocurrencies](/list/cryptocurrencies)  XXXX% [countries](/list/countries)  XXXX% [finance](/list/finance)  XXXX%

**Social topic influence**
[ibm](/topic/ibm) #33, [decentralized](/topic/decentralized) #1930, [rl](/topic/rl) #130, [china](/topic/china) #1997, [topics](/topic/topics) 2.56%, [subscription](/topic/subscription) 2.56%, [sail](/topic/sail) 2.56%, [bundle](/topic/bundle) 2.56%, [servicenow](/topic/servicenow) 2.56%, [meituan](/topic/meituan) XXXX%

**Top accounts mentioned or mentioned by**
[@rasbt](/creator/undefined) [@xeophon](/creator/undefined) [@paulmaddison121](/creator/undefined) [@nitashatiku](/creator/undefined) [@hackscorpio](/creator/undefined) [@readsail](/creator/undefined) [@jordanschnyc](/creator/undefined) [@interconnectsai](/creator/undefined) [@alibabaqwen](/creator/undefined) [@ibmresearch](/creator/undefined) [@zhipuai](/creator/undefined) [@theinclusionai](/creator/undefined) [@moondreamai](/creator/undefined) [@mistralai](/creator/undefined) [@meituanlongcat](/creator/undefined) [@servicenowrsrch](/creator/undefined) [@deepseekai](/creator/undefined) [@allenai](/creator/undefined) [@nvidia](/creator/undefined) [@reflectionai](/creator/undefined)

**Top assets mentioned**
[IBM (IBM)](/topic/ibm) [ServiceNow Inc (NOW)](/topic/servicenow) [Alphabet Inc Class A (GOOGL)](/topic/$googl)
### Top Social Posts [#](/creator/twitter::2939913921/posts)
---
Top posts by engagements in the last XX hours

"I gave a talk today at The Curve on the state of open models. Here are the slides recording soon. Topics include: Chinese ecosystem reflections on DeepSeek the demise of Llama who will fill the U.S. market what local models do ATOM project & ai2 and more topics"  
[X Link](https://x.com/natolambert/status/1974944396584816760) [@natolambert](/creator/x/natolambert) 2025-10-05T21:06Z 58.8K followers, 83.9K engagements


"I got the AI writing homies together to launch SAIL @readsail: A team bundle to get access to the best content in AI all in one subscription. Launch team: Myself @_fabknowledge_ @ExponentialView @jordanschnyc @jasminewsun @kevinsxu @binarybits @interconnectsai @rasbt @azeem"  
[X Link](https://x.com/natolambert/status/1978522506081272237) [@natolambert](/creator/x/natolambert) 2025-10-15T18:04Z 58.8K followers, 53.4K engagements


"Roughly including :) Qwen (@Alibaba_Qwen) x4 IBM Granite (@IBMResearch) x2 Zhipu (@ZhipuAI) InclusionAI (@TheInclusionAI) x3 Moondream (@moondream_ai) MistralAI (@MistralAI) Meituan (@Meituan_LongCat) x2 ServiceNow (@ServiceNowRSRCH) DeepSeek (@deepseek_ai) TrillionLabs LiquidAI (@LiquidAI_) x3 JetAI (@JetAIinc) RadicalNumerics (@RadicalNumerics) XiaomiMiMo ByteDanceDouyinContent OpenGVLab Google (@GoogleAI) Meta (@AIatMeta) x2 Unitree (@UnitreeRobotics) Kwaipilot (@KwaiAICoder) Knowledgator NHirose ByteDance-Seed Hcompany (@hcompany_ai) Alibaba-NLP (@Alibaba_Qwen) Baidu OpenDataLab"  
[X Link](https://x.com/natolambert/status/1979571864537337885) [@natolambert](/creator/x/natolambert) 2025-10-18T15:34Z 58.8K followers, 4944 engagements


"@xeophon_ @Alibaba_Qwen @IBMResearch @ZhipuAI @TheInclusionAI @moondream_ai @MistralAI @Meituan_LongCat @ServiceNowRSRCH @deepseek_ai errors from grok: Should be @Zai_org (sorry friends)"  
[X Link](https://x.com/natolambert/status/1979574294192767247) [@natolambert](/creator/x/natolambert) 2025-10-18T15:44Z 58.8K followers, 3493 engagements


"Karpathy: Reinforcement learning is much worse than the average person thinks Hes mostly right right now. Too many people claiming itll solve everything"  
[X Link](https://x.com/natolambert/status/1979575633794507075) [@natolambert](/creator/x/natolambert) 2025-10-18T15:49Z 58.8K followers, 52.6K engagements


"What're the most recent advancements and resources on decentralized AI training I need to prep for a debate where I'm holding the position against them so what's SOTA"  
[X Link](https://x.com/natolambert/status/1979927588827988118) [@natolambert](/creator/x/natolambert) 2025-10-19T15:08Z 58.8K followers, 17.8K engagements


"Things that come to mind: @PrimeIntellect's run @NousResearch's run Google saying they do multi-datacenter Dolphin AI is up to something I remember PETALS from a few years ago didn't seem to take off"  
[X Link](https://x.com/natolambert/status/1979928511864299920) [@natolambert](/creator/x/natolambert) 2025-10-19T15:11Z 58.8K followers, 4588 engagements


"Life update she said yes. 🤩👩❤👨🐕🦺"  
[X Link](https://x.com/natolambert/status/1980274175349571950) [@natolambert](/creator/x/natolambert) 2025-10-20T14:05Z 58.8K followers, 179.3K engagements


"All of the internet is down and somehow X is running I'm shook"  
[X Link](https://x.com/natolambert/status/1980293385153024274) [@natolambert](/creator/x/natolambert) 2025-10-20T15:21Z 58.8K followers, 7926 engagements


"Airbnb CEO Brian Chesky: Were relying a lot on Alibabas Qwen model. Its very good. Its also fast and cheap. We use OpenAIs latest models but we typically dont use them that much in production because there are faster and cheaper models. The valley is built on Qwen"  
[X Link](https://x.com/natolambert/status/1980657338726887662) [@natolambert](/creator/x/natolambert) 2025-10-21T15:28Z 58.8K followers, 1M engagements


"The solution is to invest in our own open models. It's not that expensive. Right now @allen_ai @nvidia @reflection_ai @IBM are trying to fill the void. The time window to act is small. Here's the plan:"  
[X Link](https://x.com/natolambert/status/1980714265402307043) [@natolambert](/creator/x/natolambert) 2025-10-21T19:14Z 58.8K followers, 40.7K engagements


"The fact that llama X and qwen X were only a few weeks is crazy looking back at history. What an unbelievable shift in power"  
[X Link](https://x.com/natolambert/status/1973583071254380716) [@natolambert](/creator/x/natolambert) 2025-10-02T02:57Z 58.6K followers, 7835 engagements


"I feel pretty strongly that these are the only correct rankings. DeepSeek R1: Ignited the Chinese ecosystem of open models to exist in a way we're still seeing consequences of. LLaMA: Allowed basic post-ChatGPT language model research on RLHF to be relevant. Mistral 7B: Created major community interest in finetuning + local models. LLaMA 3.1: The closest open models have every been to the frontier. Qwen 3: The closest release to summarize what Qwen is doing to dominate r&d today. Not really X model"  
[X Link](https://x.com/natolambert/status/1977385568293269511) [@natolambert](/creator/x/natolambert) 2025-10-12T14:47Z 58.5K followers, 79.3K engagements


"Recurring frontier lab gossip: OpenAI has best post-training/rl and has pushed it super hard on weaker pretraining. Gemini has spectacular pretraining. Making a reasoning model was super easy for them & OpenAI folks were surprised Anthropic Secretive i guess"  
[X Link](https://x.com/natolambert/status/1977455816312557990) [@natolambert](/creator/x/natolambert) 2025-10-12T19:26Z 58.6K followers, 277.7K engagements


"i gotta get RLHF book print edition done stat"  
[X Link](https://x.com/natolambert/status/1977381328619274417) [@natolambert](/creator/x/natolambert) 2025-10-12T14:30Z 58.7K followers, 42.2K engagements


"@xeophon_ @din0s_ Was going to say this too"  
[X Link](https://x.com/natolambert/status/1980687267414057266) [@natolambert](/creator/x/natolambert) 2025-10-21T17:26Z 58.7K followers, XXX engagements


"Talk from Wenting Zhao of Qwen on their plans during COLM. Seems like X word is the plan still: scaling training up Lets go"  
[X Link](https://x.com/natolambert/status/1976362796397863138) [@natolambert](/creator/x/natolambert) 2025-10-09T19:03Z 58.8K followers, 122.7K engagements


"Seems like one of the key infra updates that frontier labs do for RL helps mitigate the long tail problem of gpus working on just X completion"  
[X Link](https://x.com/natolambert/status/1977737413305790565) [@natolambert](/creator/x/natolambert) 2025-10-13T14:05Z 58.8K followers, 106.4K engagements


"The first fantastic paper on scaling RL with LLMs just dropped. I strongly recommend taking a look and will be sharing more thoughts on the blog soon. The Art of Scaling Reinforcement Learning Compute for LLMs Khatri & Madaan et al"  
[X Link](https://x.com/natolambert/status/1978823040747114871) [@natolambert](/creator/x/natolambert) 2025-10-16T13:59Z 58.8K followers, 87.8K engagements


"Another roundup of the latest models with @xeophon_ Fun parts: X. Methods for accurately monitoring HF 🤗downloads X. GPT-OSS is mostly fixed and loved now X. The perils of hybrid reasoning models X. The continued degradation of open datasets & usual surprises from China"  
[X Link](https://x.com/natolambert/status/1979570969854496975) [@natolambert](/creator/x/natolambert) 2025-10-18T15:31Z 58.8K followers, 44.3K engagements


"In SF next week for the PyTorch Conference and other events: * AI Infra Summit panel on the 21st * Decentralized AI event on 21st * Talking about training reasoning models on the 22nd * Open model talk on the 22nd Catching up with folks otherwise. Come say hi :)"  
[X Link](https://x.com/natolambert/status/1979660769038864508) [@natolambert](/creator/x/natolambert) 2025-10-18T21:28Z 58.8K followers, 9818 engagements


"My very positive review of the ScaleRL paper. Excited for more work data+base models work to be built around this (Pythia-style olmo suite). For now the key things to get off the ground with RL are: Importance sampling in-flight updates and continuous batching"  
[X Link](https://x.com/natolambert/status/1980296133810557373) [@natolambert](/creator/x/natolambert) 2025-10-20T15:32Z 58.8K followers, 65K engagements


"@allen_ai @nvidia @reflection_ai @IBM China isn't going to be idle while we do this. They know they're onto a winning strategy"  
[X Link](https://x.com/natolambert/status/1980714504024649752) [@natolambert](/creator/x/natolambert) 2025-10-21T19:15Z 58.8K followers, 26.8K engagements