Carta Thomas
@CartaThomas2
Followers
83
Following
14
Media
11
Statuses
60
Ph.D student at INRIA in the @FlowersINRIA. I am working on the how language and RL interact
Joined February 2023
RT @edwardfhughes: The automation of innovation is within reach! Delighted that my @raais talk is now available for anyone to watch, alongsโฆ.
0
16
0
RT @edwardfhughes: Human ideation beats AI ideation when measured on execution outcomes: There's a clear path to fโฆ.
0
8
0
RT @LorisGaven: ๐ Join our MAGELLAN talk on July 2!. We'll explore how LLM agents can monitor their own learning progress and choose what tโฆ.
0
3
0
RT @ClementRomac: ๐ฐ Check out the full paper here: Don't hesitate to reach out if you want to discuss WorldLLMs orโฆ.
0
1
0
RT @ClementRomac: I'll be at RLDM this week to present our new paper: WorldLLM ๐. In the same spirit as our previous works โ e.g. GLAM, MAGโฆ.
0
6
0
RT @hardmaru: New Paper: Continuous Thought Machines ๐ง . Neurons in brains use timing and synchronization in the way that they compute, butโฆ.
0
571
0
RT @sherryyangML: Checkout Generative Hierarchical Materials Search (GenMS) โ a framework for generating crystal structures from natural laโฆ.
0
28
0
RT @pyoudeyer: Generative AI is a cultural transmission technology: .it plays a growing role in generation, selection and transmission ofโฆ.
0
6
0
RT @nicolas__yax: ๐ฅOur paper PhyloLM got accepted at ICLR 2025 !๐ฅ.In this work we show how easy it can be to infer relationship between LLMโฆ.
0
13
0
RT @SakanaAILabs: ้ฒๅใใAIใตใคใจใณใใฃในใใใThe AI Scientist-v2ใใๅ
ฌ้. AIใๅฎๅ
จ่ชๅใง็งๅญฆ่ซๆใ็ๆใใAIใตใคใจใณใใฃในใใฎๆน่จ็ใThe AI Scientist-v2ใๅ
ฌ้ใใพใใใ. ่ซๆ๏ผ.
0
70
0
RT @SakanaAILabs: Introducing The AI Scientist-v2, which produced the 1st fully AI-generated paper to pass peer review at a workshop levelโฆ.
0
245
0
RT @_tom_bush: ๐ค !! Model-free agents can internally plan !! ๐ค. In our ICLR 2025 paper, we interpret a model-free RL agent and show that itโฆ.
0
46
0
RT @_zzach_: I used ADTool to conduce the exploration @itsjoshsurely @PapalopuluLab @v_mochulska @HiromuneEto @Sonnโฆ.
0
1
0
RT @a_lidayan: ๐จOur new #ICLR2025 paper presents a unified framework for intrinsic motivation and reward shaping: they signal the value ofโฆ.
0
32
0
๐งญMAGELLAN is built on on many works that use of LP to drive automatic curriculum learning e.g. by @_rockt @egrefen @jeffclune @jennyzhangzt @ingkanit @RPortelas @forestier_seb @nishanthkumar23 @tomssilver @tambetm @johnschulman2 @RyanSullyvan.
0
0
3
Thanks to @LorisGaven @ClementRomac for the fun time doing research on this topic and huge thanks also to, @cedcolas @SylvainLamprier, @OlivierSigaud and @pyoudeyer for their supervision!!.
4
0
1
๐๐: ๐๐๐๐ฉ๐ญ๐๐ญ๐ข๐จ๐ง ๐ญ๐จ ๐๐ฏ๐จ๐ฅ๐ฏ๐ข๐ง๐ ๐๐จ๐๐ฅ ๐๐ฉ๐๐๐๐ฌ.We replaced the ๐๐ง๐ญ๐ข๐ซ๐ ๐ ๐จ๐๐ฅ ๐ฌ๐ฉ๐๐๐ with unseen goals from the same categories. ๐งญMAGELLAN generalized LP and retained exceptional performanceโmatching baselines that rely on human expertise! ๐โจ
1
0
1
๐๐: ๐๐๐ง๐๐ซ๐๐ฅ๐ข๐ณ๐๐ญ๐ข๐จ๐ง.At the end of training, ๐งญMAGELLAN has ๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐๐ ๐ญ๐ก๐ ๐ ๐จ๐๐ฅ ๐๐ฆ๐๐๐๐๐ข๐ง๐ ๐ฌ๐ฉ๐๐๐, consistently ๐ฉ๐ซ๐๐๐ข๐๐ญ๐ข๐ง๐ success probability ๐๐จ๐ซ ๐ฎ๐ง๐ฌ๐๐๐ง ๐ ๐จ๐๐ฅ๐ฌ, a key step toward scalable open-ended learning!
1
0
1
๐๐: ๐๐ฎ๐ซ๐ซ๐ข๐๐ฎ๐ฅ๐ฎ๐ฆ ๐๐๐๐ซ๐ง๐ข๐ง๐ .๐งญMAGELLAN autonomously discovers goal families (โ๐ฟ๐ฎ๐ฆ) across ๐๐๐ค ๐ ๐จ๐๐ฅ๐ฌ, performing on par with expert knowledge-augmented baselinesโbut ๐ฐ๐ข๐ญ๐ก๐จ๐ฎ๐ญ ๐ซ๐๐ช๐ฎ๐ข๐ซ๐ข๐ง๐ ๐ฉ๐ซ๐๐๐๐๐ข๐ง๐๐ ๐ ๐จ๐๐ฅ ๐๐ฅ๐ฎ๐ฌ๐ญ๐๐ซ๐ฌ! ๐
1
0
1