Aditi Mavalankar
@aditimavalankar.bsky.social
Research Scientist at DeepMind working on Gemini Thinking
Reposted by Aditi Mavalankar
Where do some of Reinforcement Learning's great thinkers stand today?
Find out! Keynotes of the RL Conference are online:
www.youtube.com/playlist?lis...
Wanting vs liking, Agent factories, Theoretical limit of LLMs, Pluralist value, RL teachers, Knowledge flywheels
(guess who talked about which!)
Find out! Keynotes of the RL Conference are online:
www.youtube.com/playlist?lis...
Wanting vs liking, Agent factories, Theoretical limit of LLMs, Pluralist value, RL teachers, Knowledge flywheels
(guess who talked about which!)
August 27, 2025 at 12:46 PM
Where do some of Reinforcement Learning's great thinkers stand today?
Find out! Keynotes of the RL Conference are online:
www.youtube.com/playlist?lis...
Wanting vs liking, Agent factories, Theoretical limit of LLMs, Pluralist value, RL teachers, Knowledge flywheels
(guess who talked about which!)
Find out! Keynotes of the RL Conference are online:
www.youtube.com/playlist?lis...
Wanting vs liking, Agent factories, Theoretical limit of LLMs, Pluralist value, RL teachers, Knowledge flywheels
(guess who talked about which!)
On my way to #ICML2025 to present our algorithm that strongly scales with inference compute, in both performance and sample diversity! 🚀
Reach out if you’d like to chat more!
Reach out if you’d like to chat more!
July 13, 2025 at 12:26 PM
On my way to #ICML2025 to present our algorithm that strongly scales with inference compute, in both performance and sample diversity! 🚀
Reach out if you’d like to chat more!
Reach out if you’d like to chat more!
Reposted by Aditi Mavalankar
New side project!
assayer: A simple Python-RQ based tool to automatically monitor and evaluate ML model checkpoints offline during training.
assayer: A simple Python-RQ based tool to automatically monitor and evaluate ML model checkpoints offline during training.
June 15, 2025 at 10:27 PM
New side project!
assayer: A simple Python-RQ based tool to automatically monitor and evaluate ML model checkpoints offline during training.
assayer: A simple Python-RQ based tool to automatically monitor and evaluate ML model checkpoints offline during training.
Reposted by Aditi Mavalankar
Ever thought of joining DeepMind's RL team? We're recruiting for a research engineering role in London:
job-boards.greenhouse.io/deepmind/job...
Please spread the word!
job-boards.greenhouse.io/deepmind/job...
Please spread the word!
Research Engineer, Reinforcement Learning
London, UK
job-boards.greenhouse.io
May 22, 2025 at 3:11 PM
Ever thought of joining DeepMind's RL team? We're recruiting for a research engineering role in London:
job-boards.greenhouse.io/deepmind/job...
Please spread the word!
job-boards.greenhouse.io/deepmind/job...
Please spread the word!
Accepted to #ICML2025
See you in Vancouver!
See you in Vancouver!
Excited to share our recent work, AuPair, an inference-time technique that builds on the premise of in-context learning to improve LLM coding performance!
arxiv.org/abs/2502.18487
🧵
arxiv.org/abs/2502.18487
🧵
AuPair: Golden Example Pairs for Code Repair
Scaling up inference-time compute has proven to be a valuable strategy in improving the performance of Large Language Models (LLMs) without fine-tuning. An important task that can benefit from additio...
arxiv.org
May 2, 2025 at 10:35 AM
Accepted to #ICML2025
See you in Vancouver!
See you in Vancouver!
Reposted by Aditi Mavalankar
When faced with a challenge (like debugging) it helps to think back to examples of how you've overcome challenges in the past. Same for LLMs!
The method we introduce in this paper is efficient because examples are chosen for their complementarity, leading to much steeper inference-time scaling! 🧪
The method we introduce in this paper is efficient because examples are chosen for their complementarity, leading to much steeper inference-time scaling! 🧪
Excited to share our recent work, AuPair, an inference-time technique that builds on the premise of in-context learning to improve LLM coding performance!
arxiv.org/abs/2502.18487
🧵
arxiv.org/abs/2502.18487
🧵
AuPair: Golden Example Pairs for Code Repair
Scaling up inference-time compute has proven to be a valuable strategy in improving the performance of Large Language Models (LLMs) without fine-tuning. An important task that can benefit from additio...
arxiv.org
March 20, 2025 at 10:23 AM
When faced with a challenge (like debugging) it helps to think back to examples of how you've overcome challenges in the past. Same for LLMs!
The method we introduce in this paper is efficient because examples are chosen for their complementarity, leading to much steeper inference-time scaling! 🧪
The method we introduce in this paper is efficient because examples are chosen for their complementarity, leading to much steeper inference-time scaling! 🧪
Excited to share our recent work, AuPair, an inference-time technique that builds on the premise of in-context learning to improve LLM coding performance!
arxiv.org/abs/2502.18487
🧵
arxiv.org/abs/2502.18487
🧵
AuPair: Golden Example Pairs for Code Repair
Scaling up inference-time compute has proven to be a valuable strategy in improving the performance of Large Language Models (LLMs) without fine-tuning. An important task that can benefit from additio...
arxiv.org
March 17, 2025 at 11:16 AM
Excited to share our recent work, AuPair, an inference-time technique that builds on the premise of in-context learning to improve LLM coding performance!
arxiv.org/abs/2502.18487
🧵
arxiv.org/abs/2502.18487
🧵
Reposted by Aditi Mavalankar
Are there limits to what you can learn in a closed system? Do we need human feedback in training? Is scale all we need? Should we play language games? What even is "recursive self-improvement"?
Thoughts about this and more here:
arxiv.org/abs/2411.16905
Thoughts about this and more here:
arxiv.org/abs/2411.16905
Boundless Socratic Learning with Language Games
An agent trained within a closed system can master any desired capability, as long as the following three conditions hold: (a) it receives sufficiently informative and aligned feedback, (b) its covera...
arxiv.org
November 28, 2024 at 4:01 PM
Are there limits to what you can learn in a closed system? Do we need human feedback in training? Is scale all we need? Should we play language games? What even is "recursive self-improvement"?
Thoughts about this and more here:
arxiv.org/abs/2411.16905
Thoughts about this and more here:
arxiv.org/abs/2411.16905