Reshinth

How to define Diversity in the context of CodeLMs and Programming Languages ? 1. Diversity is positively correlated with Performance in solving a problem. 2. Shortcomings of diversity in small codeLMs. 3. Code Embedding models don't capture semantics. reshinthadithyan.github.io/blog/2023/code…

🎬 Introducing Stable Cinemetrics, to be presented at NeurIPS 2025. We present the first taxonomy of professional controls to systematically study and control video generative models through the lens of filmmaking. Interactive webpage with paper link: stable-cinemetrics.github.io 🧵

@tylerangert Happy Birthday 🎉

My forever fun fact: my birthday is the calendar emoji 📅

🙋‍♂️ Can RL training address model weaknesses without external distillation? 🚀 Please check our latest work on RL for LLM reasoning! 💯 TL;DR: We propose augmenting RL training with synthetic problems targeting model’s reasoning weaknesses. 📊Qwen2.5-32B: 42.9 → SwS-32B: 68.4

@evanthebouncy Congrats Evan.

I've recently started my job as an asst professor at NTU, Singapore. If you are ever in town come say hi :)

With CodeLMs scaling actually solved models intrinsically learning internal structural syntactical & semantic information.

Open AI gave a talk on writing software through specs today. I thought it was my little secret, but seems like quite a few smart builders in the space have also found it's a useful approach. Now that the secrets out joshuapurtell.com/posts/spec_eng/

Our Infinibranch Sandboxes power @huggingface OpenR1's code-based rewards for training LLMs with GRPO. From multi-file complex evaluation pipelines for IOI problems in C++, to Jupyter execution for Python: we evaluate in seconds, not minutes github.com/huggingface/op…

@scychan_brains Resonates a lot. I myself found stuck in such scenario late last year. Talking with collaborators and colleagues helped a lot. Thanks a lot for writing. ♥️

Some years ago, I got trapped in a Massive Trough of Imposter Syndrome. It took more than a year to dig myself out of it, but the following framework really helped me. It feels a bit vulnerable to share, but I hope it might help a few others too! A short thread 🧵🙂

Does anyone else remember a year or so ago when all the AI podcasts did AGI debates? What an awful time that was, glad it’s over

@FraserGreenlee @arcprize Am I understanding it right that you're describing program synthesis ?

Wonder if there's room for an @arcprize game where you gradually build up an abstraction library to solve tasks of increasing difficulty. Bonus points for using less code.

🧵We just released the #1 open-source agent on the SWE-bench Verified leaderboard by assembling the best of Claude Sonnet 3.7 and O1. Open-source repo here: github.com/augmentcode/au… Here's how we achieved 65.4% success rate on the hardest coding benchmark in the industry: 🧠👇

OpenAI's CUA plays Pokemon in the multiverse with Infinibranch by Morph Cloud choose every starter: no more pesky 'decisions'

ML Twitter lately.

How important is the quality, diversity, and complexity (QDC) of synthetic data for LLM performance? What effect does QDC data composition have on self-improvement? We just released a comprehensive survey discussing these questions (and many more) 🧵

As R&D staff @answerdotai, I work a lot on boosting productivity with AI. A common theme that always comes up is the combination of human+AI. This combination proved to be powerful in our new project ShellSage, which is an AI terminal buddy that learns and teaches with you. A 🧵

New Anthropic research: Adding Error Bars to Evals. AI model evaluations don’t usually include statistics or uncertainty. We think they should. Read the blog post here: anthropic.com/research/stati…

@ncooper57 @fastdotai Congrats Nathan <3

I'm so excited to be working on this new course from @fastdotai ! Education has always been a huge driving factor in my life. It is surreal that I'm getting to do this as part of my job. Really looking forward to working with students again 🤓