Great to hear that OpenAI has confirmed that the startup Weco AI, co-founded by my former intern Zhengyao Jiang, has the best Machine Learning Engineer Agent in the world😀in their MLE-Bench. Congrats!

Amazing progress Yuxiang (Jimmy) Wu and Zhengyao Jiang, and great to see the impact of "agent scaffolding" given a base model.

We’re hiring a full-time Frontend Engineer! Join us to build next-gen apps that bring AI to life through natural language. If you’re passionate about product and AI, apply now! #hiring #frontend #AI linkedin.com/jobs/view/4069…

How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this: Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢 🧵⬇️

How close are current AI agents to automating AI R&D? Our new ML research engineering benchmark (RE-Bench) addresses this question by directly comparing frontier models such as Claude 3.5 Sonnet and o1-preview with 50+ human experts on 7 challenging research engineering tasks.

AIDE was built for tabular Machine Learning and optimized for GPT-4. It surprised me by generalizing to new models (o1) & deep learning tasks by OpenAI MLE-Bench. RE-Bench now shows it scaling to cutting-edge AI research, this is mind-blowing!

AIDE by Weco AI's Dominik Schmidt Zhengyao Jiang and Yuxiang (Jimmy) Wu strikes again!

🚨 New Paper 🚨 Can LLMs perform latent multi-hop reasoning without exploiting shortcuts? We find the answer is yes – they can recall and compose facts not seen together in training or guessing the answer, but success greatly depends on the type of the bridge entity (80%+ for

The state-of-the-art MLE agent AIDE now has a local Web UI. Connect it with your favourite LLM and start playing now!

Mitigating racial bias from LLMs is a lot easier than removing it from humans! Can’t believe this happened at the best AI conference NeurIPS Conference We have ethical reviews for authors, but missed it for invited speakers? 😡

Over the past few months, the Weco AI team has been hard at work building Weco AI Functions—a platform that simplifies adding and optimizing AI features with just a function call. My favorite part? Effortless A/B testing and versioning. You can compare multiple LLMs

Nice experience😀. Define a function with natural language, and the function call is available to you immediately anywhere. "What you think immediately becomes what you get" 🚀🚀

We spoke with Laura Ruis from Cohere For AI and UCL about her paper "Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models" where she demonstrated an interesting gap between retrieval and reasoning queries in LLMs indicating the presence of synthesised

The entire internet era has been about transmitting information. AI will take computer science to the next level, by generating new knowledge through trial and error. LLMs following the scientific methodology can already automate R&D. Check the paper of AIDE!

I used to spend weeks in trial-and-error loops building deep learning models, until we built AIDE to handle that work for us. Now I can tackle more than 20 ML problems at once and train 1,000+ models in parallel. It’s incredibly empowering! See how we’re rethinking machine

If you are at ICLR and interested in AI-driven scientific discovery, make sure to chat with Zhengyao Jiang

Love this project: nanoGPT -> recursive self-improvement benchmark. Good old nanoGPT keeps on giving and surprising :) - First I wrote it as a small little repo to teach people the basics of training GPTs. - Then it became a target and baseline for my port to direct C/CUDA

Thrilled to see Weco AI's AIDE used in Meta's work and big congrats to Minqi Jiang Bingchen Zhao Despoina Magka. It's a truly exciting time to work on Recursive Self-Improvement.

Seed secured, now the real experiments begin Weco AI