🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

We put a basic demo Colab for Direct Ascent Synthesis (DAS) (arxiv.org/abs/2502.07753) up on Github: github.com/stanislavfort/… You should be able to 1) generate images from text, 2) run "style" transfer, and 3) reconstruct images from CLIP embeddings in minutes on an A100

I'm trying to build a quick and dirty semantic search over a repository. What are the best embedding models out there right now? I'm a bit out of the loop. Or should I be doing something else than comparing embedding vectors of a query to code chunks?

This is still genuinely surprising: I use LLMs all the time to read research papers, to write code, to brainstorm ideas, and rarely see any issues at all + get a huge amount of productivity gain from them. Yet to many they're just valueless hallucinators. What's going on here?

New preprint: we evaluated LLMs in a 3-party Turing test (participants speak to a human & AI simultaneously and decide which is which). GPT-4.5 (when prompted to adopt a humanlike persona) was judged to be the human 73% of the time, suggesting it passes the Turing test (🧵)

This is very much a landmark paper. An empirical demonstrations that humans are able to learn concepts from super-human AIs that before that were inaccessible to them

Isn't the Strong Model Collapse paper basically impossible to be correct since synthetic data is a huge part of frontier model training already? > results show that even the smallest fraction of synthetic data (e.g., as little as 1% [...]) can still lead to model collapse ???

My impression was that this was relatively widely accepted since the landmark paper on translation without parallel corpora in 2017 titled "Word Translation Without Parallel Data"? It's got >1k citations and spawned a small subfield

A good summary of a sadly pretty suboptimal paper that went viral recently

I have also been seeing a persistent hallucinatory citation of a paper I am allegedly the first author of, which, despite that, sadly doesn't exist

I'll be in Cambridge UK for a few days, DM me if you'd like to meet up