Tianlong Xu

@_philschmid Would this be easily scalable to multi-modal case? I am wondering if we could see such aha moments in image understanding by teaching it to think deep.

Mini-R1: Reproduce Deepseek R1 „aha moment“ a RL tutorial! Recreate an RL "aha moment" using Group Relative Policy Optimization (GRPO) and train an open model using reinforcement learning to teach it self-verification and search abilities all on its own to solve the Countdown Game. philschmid.de/mini-deepseek-…

@jiayi_pirate Hi Jiayi, this is really inspiring - is there a multi-modal version out there, or do you plan to create something soon?

We reproduced DeepSeek R1-Zero in the CountDown game, and it just works Through RL, the 3B base LM develops self-verification and search abilities all on its own You can experience the Ahah moment yourself for < $30 Code: github.com/Jiayi-Pan/Tiny… Here's what we learned 🧵

As a quick snippet of the ideas in these two papers: one is to use LLMs to detect the root cause of a student who has made an error from their digital drafts. And the other is to use multi-agents to perform alignments between math questions and knowledge concepts.

Check out our two papers accepted by #AAAI2025. "AI-Driven Virtual Teacher for Enhanced Educational Efficiency: Leveraging Large Pretrain Models for Autonomous Error Analysis and Correction" and "Knowledge Tagging with Large Language Model based Multi-Agent System" .

The links are arxiv.org/abs/2409.09403 and arxiv.org/abs/2409.08406

@joaomdmoura @langchain so impressive😀

Fifty years ago, Apollo astronauts first landed on the Moon. What better way to celebrate than constructing tiny versions of some of the mission's iconic vehicles and scientists? Sign up for Nature Briefing for a chance to win (current readers can enter and win too) 🚀

@southerndsc very exciting event. Enjoying it. #SDSC19 #SDSC2019

#SDSC2019#

@realDonaldTrump :)