🔥 Excited to share our new work on Neural Flow Diffusion Models — a general, end-to-end, simulation-free framework that works with an arbitrary noising processes and even enables learning them! 📜: arxiv.org/abs/2404.12940 🧵 1/11

There's a big difference between solving a problem from first principles vs applying a solution template you previously memorized. It's like the difference between a senior software engineer and a script kiddie that can't code. A script kiddie that has a gigantic bank of scripts

sonnet 3.5 is a great model for coding, but it is just way too eager to reply in bullet points on basic text prompts

Are you going to #ICML2024? These top faculty and researchers from MBZUAI will be there presenting their latest pioneering work! We are excited to announce that 25 papers from MBZUAI have been accepted for presentation at the 41st International Conference on Machine Learning

He is not wrong. Genuinely enjoyed this part of the discussion about current hype cycle in ML research.

Med42-v2 A Suite of Clinical LLMs discuss: huggingface.co/papers/2408.06… Med42-v2 introduces a suite of clinical large language models (LLMs) designed to address the limitations of generic models in healthcare settings. These models are built on Llama3 architecture and fine-tuned

Genomic Foundationless Models: Pretraining Does Not Promise Performance 1. This study challenges the paradigm of pretraining in Genomic Foundation Models (GFMs), revealing that randomly initialized models often match or surpass pretrained ones in fine-tuning tasks. 2. Despite

New preprint claims that most existing DNA language models perform just as well with random weights, suggesting that pretraining does nothing (Mistral & DNABERT-2 look like exceptions). We need better DNA language models.

How different are the outputs of various LLMs, and in what ways do they differ? Turns out, very very different, up to the point that a text encoding classifier could tell the source LLM with 97% accuracy. This is classifying text generated by LLMs, between ChatGPT, Claude,

Introducing BioFM, a biologically-informed GFM that: ✅ Outperforms all small GFMs (265M params, trained on just 50 genomes) ✅ Beats Evo2-7B (variant embeddings), Enformer (expression), SpliceTransformer (sQTL). No brute-force scaling-just smarter tokenization.