juri

Unless I’m mistaken, common versions of ChatGPT and Claude use conversations for training—meaning they can pick up ideas in unpublished work and pass them on to other users; laundering idea plagiarism and creating “scooping” dynamics without any of the people involved knowing it.

I never want to review an ML paper ever again. Most of the good ML researchers go work in industry instead of submitting public papers, so ML conference papers are adversely selected and are on average terrible. This mood brought to you by: having to review ICML.

Controversial take: you don't need any of this. LLMs have gone through a lot of training already, so there ought to be a better method to turn them into extremely good embedding models. This is what my group has been working on. LLM2Vec is one such idea. We have some exciting developments recently where LLMs themselves can generate superior embeddings with zero changes to the LLM. Stay tuned!

Your LLM already knows the answer. Why is your embedding model still encoding the question? 🚨Introducing LLM2Vec-Gen: your frozen LLM generates the answer's embedding in a single forward pass — without ever generating the answer. Not only that, the frozen LLM can decode the embedding back into text. 🏆 SOTA self-supervised embeddings 🛡️ Free transfer of instruction-following, safety, and reasoning

there, I said it. diffusion LLMs are the future! I'll be back in a couple of years to collect my "I told you so" award.

Something weird just happened to me. A new paper was linked with my profile in research gate. I am listed as an author of this paper. However, I have not written this paper, and I don't know this paper.