Andrew Wang

Our paper on influence functions titled Better TDA via Better Inverse Hessian-Vector Products will be presented at #neurips25 San Diego on Friday from 4:30 PM - 7:30 PM in Exhibit hall C,D,E (Poster 3907) Work with my fantastic colleagues @_elinguyen, @Runshi_Yang , @juhan_bae, @SheilaMcIlraith, @RogerGrosse!

Proud to introduce Llemma, the first models trained on OpenWebMath (part of the 55B mathematical tokens in ProofPile II). Llemma serves as a platform for future research on quantitative reasoning and is a very powerful base model for mathematical tasks!

Should you let LMs control your email? terminal? bank account? or even your smart home?🤔 🔥Introducing ToolEmu for identifying risks associated with LM agents at scale! 🛠️Featuring LM-emulation of tools & automated realistic risk detection 🚨GPT4 is risky in 40% of our cases!

Identifying the Risks of LM Agents with an LM-Emulated Sandbox Presents a framework that uses a LM to emulate tool execution and enables scalable testing of LM agents against a diverse range of tools and scenarios proj: toolemu.com abs: arxiv.org/abs/2309.15817

This paper was a central contribution of @RToroIcarte 's PhD thesis at U of T, and done in collaboration with Toryn Klassen, @RickValenzano, and supervisor, @SheilaMcIlraith. Early results were first published at ICML18 and AAMAS18. We are proud to have received this recognition.

Meet STEVE-1, an instructable generative model for Minecraft. STEVE-1 follows both text and visual instructions and acts on raw pixel inputs with keyboard and mouse controls. Best of all - it only cost $60 to train! w/ @Shalev_lif @SirrahChan @jimmybajimmyba @SheilaMcIlraith

Early LLM induced multibillion dollar casualty. What is the value of text-based content, if it can be paraphrased and accessed for free? bloomberg.com/news/articles/…

Excited to share Boosted Prompt Ensembles! Inspired by boosting algorithms, we propose an algorithm that automatically grows an ensembles of prompts to cover a target problem space. Led by @silviupitis and with @andrew_wang10 and @jimmybajimmyba

What are boosted prompt ensembles, and how can they be used? New paper released yesterday significantly improves output quality through automating n-shot learning Paper breakdown with useful takeaways 📃🧵👇 arxiv.org/abs/2304.05970

Generative AI will likely disrupt a number of industries. But as of now, it seems to me the biggest disruption is to existing AI researchers.

Apple launched the app store in 2008. This is the 2023 version.