Promising. Everyone should hope that we can throw away tokenization in LLMs. Doing so naively creates (byte-level) sequences that are too long, so the devil is in the details. Tokenization means that LLMs are not actually fully end-to-end. There is a whole separate stage with

The first RNN in transformers! 🤯 Announcing the integration of RWKV models in transformers with BlinkDL and RWKV community! RWKV is an attention free model that combines the best from RNNs and transformers. Learn more about the model in this blogpost: huggingface.co/blog/rwkv

No one person has all the expertise needed to address AI safety.

JUST IN: Falcon 180B is here! This is one of the largest open models available today. It sits on top of the HF Open LLM Leaderboard for pre-trained models. Curious to see how this one performs on tasks like reasoning and knowledge tests. (More on this soon) They also claim

This is a fascinating deep dive into the differences between single-cell RNA-seq packages in #rstats vs #python. Not only do they produce different results, big changes in recent versions mean that older vs newer software versions also produce substantially different findings.