Andrej Karpathy joins Anthropic pretraining team and will lead a new group that uses Claude to conduct research, according to journalist Alex Heath
One post featured a satirical portrait of him as Pope Dario I.
#20Miles Brundage@MILES_BRUNDAGEBig day for those who do not wish to advance the rate of AI capabilities progress
Andrew Curran@AndrewCurran_Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.
(ref. to this, for those who are less AI company statement obsessed https://www.anthropic.com/news/core-views-on-ai-safety)
Miles Brundage@Miles_BrundageBig day for those who do not wish to advance the rate of AI capabilities progress
@scaling01 You’re right
Lisan al Gaib@scaling01Karpathy hasn't said what exactly he's going to work on, but I don't think he's getting into the weeds of optimizers or hardware or whatever I think he's going to work on something higher-level my best guess is auto-research
@scaling01 Spot on
Nicholas Joseph@nickevanjosephExcited to welcome Andrej to the Pretraining team! He'll be building a team focused on using Claude to accelerate pretraining research itself. I can’t think of anyone better suited to do it — looking forward to what we build together!
#250Richard Ngo@RICHARDMCNGOAn AGI company being “safety-focused” sometimes produces useful alignment research.
But it also focuses their capabilities researchers’ attention on the most acceleratory work.
The latter effect seems increasingly dominant.
Andrew Curran@AndrewCurran_Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.
#250Richard Ngo@RICHARDMCNGOThe more I observe this dynamic the more viscerally worried I become about Anthropic (vs other labs).
Idk how much to trust this gut-level concern. I might be treating Anthropic as outgroup and other labs as fargroup. But they just keep pushing on RSI!
#250Richard Ngo@RICHARDMCNGOPosting a few related tweets of mine below:
Richard Ngo@RichardMCNgoMany in AI safety have narrowed in on automated AI R&D as a key risk factor in AI takeover. But I'm concerned that the actions they're taking in response (e.g. publishing evals, raising awareness in labs) are very similar to the actions you'd take to accelerate automated AI R&D.
#250Richard Ngo@RICHARDMCNGOI think focusing on automated alignment research will probably make this dynamic even worse:
The same dynamic applied at OpenAI, which was focused on the shortest path to AGI back when DeepMind was still doing very broad research. And a lot of LLM scaling was driven by “safety people” like Dario under thin fig leaves which the field still pretends to believe.
Richard Ngo@RichardMCNgoI think focusing on automated alignment research will probably make this dynamic even worse:
#250Richard Ngo@RICHARDMCNGOThe same dynamic applied at OpenAI, which was focused on the shortest path to AGI back when DeepMind was still doing very broad research. And a lot of LLM scaling was driven by “safety people” like Dario under thin fig leaves which the field still pretends to believe.
we're in the x hype hiring era At least OpenAI got the OpenClaw guy
Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
#517Andrew Curran@ANDREWCURRAN_Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.
#716elie@ELIEBAKOUCH@scaling01 would be so cool
Lisan al Gaib@scaling01Karpathy hasn't said what exactly he's going to work on, but I don't think he's getting into the weeds of optimizers or hardware or whatever I think he's going to work on something higher-level my best guess is auto-research
#716elie@ELIEBAKOUCH@scaling01 lol missed this, thanks 🫡
Lisan al Gaib@scaling01@eliebakouch see two links in the thread
#839Beff (e/acc)@BEFFJEZOSKarpathy the teacher is the meta-meta-meta-learner
Andrew Curran@AndrewCurran_Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.
#980Lisan al Gaib@SCALING01Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
#980Lisan al Gaib@SCALING01shout-out to prime intellect for jumping on the auto-research wagon before Karpathy joined Anthropic to work on that
#980Lisan al Gaib@SCALING01Karpathy hasn't said what exactly he's going to work on, but I don't think he's getting into the weeds of optimizers or hardware or whatever
I think he's going to work on something higher-level
my best guess is auto-research
Lisan al Gaib@scaling01By His Holiness Pope Dario I: "Karpathy is hereby appointed Supreme RSI Architect and shall receive π/17 steradians of the future lightcone"
#980Lisan al Gaib@SCALING01Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
@eliebakouch yeah it's confirmed already
elie@eliebakouch@scaling01 would be so cool
#980Lisan al Gaib@SCALING01@eliebakouch see two links in the thread
Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
#980Lisan al Gaib@SCALING01Karpathy was literally hired for RSI
he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
Lisan al Gaib@scaling01Karpathy hasn't said what exactly he's going to work on, but I don't think he's getting into the weeds of optimizers or hardware or whatever I think he's going to work on something higher-level my best guess is auto-research
Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
#980Lisan al Gaib@SCALING01Nicholas Joseph@nickevanjosephExcited to welcome Andrej to the Pretraining team! He'll be building a team focused on using Claude to accelerate pretraining research itself. I can’t think of anyone better suited to do it — looking forward to what we build together!
Lisan al Gaib@scaling01Karpathy was literally hired for RSI he's starting a new pre-training team at Anthropic that focuses on autoresearch per Axios
#980Lisan al Gaib@SCALING01okay I guess the RSI Architect role Karpathy is taking on was too obvious
Lisan al Gaib@scaling01By His Holiness Pope Dario I: "Karpathy is hereby appointed Supreme RSI Architect and shall receive π/17 steradians of the future lightcone"
#1114alth0u🧶@ALTH0Utheyre breeding karpathy
Peter Wang 🦋@pwangWait wait I was told that AI was going to hit recursive self improvement by itself later this year… hiring Karpathy breaks the recursive step unless they’re going to actively breed or clone him
#1511Matt Turck@MATTTURCK2025: pre-training is dead!
2026:
Nicholas Joseph@nickevanjosephExcited to welcome Andrej to the Pretraining team! He'll be building a team focused on using Claude to accelerate pretraining research itself. I can’t think of anyone better suited to do it — looking forward to what we build together!