We should do a blog post on this. It could really help benchmark LLMs
RESEARCH
-
Science as Optimal Resource Allocation for Human Future
By
–
the most self-preserving resource allocation for humanity is to allocate as much as possible towards science at all times science (vs consumption, which is where the marginal dollar usually goes) is the only true investment in the future disclaimer: I was born in Los Alamos
-
Zero-shot voice transfer module restores dysarthric speech via TTS
By
–
Today we introduce a new zero-shot, cross-lingual voice transfer module that can be easily plugged into a state-of-the-art TTS system to restore the voices of speakers with dysarthric speech. See it in action at https://
youtube.com/watch?v=nOiGSq
Ekh_U
… and read more at -
On evaluating LLMs and token-scale confusions
By
–
Hard to say for LLMs because there’s no “all else being equal” — there aren’t any comparably large byte-token models. At smaller scales, I’d expect “Scunthorpe” style problems — confusions explainable as the result of spelling coincidences. I’m just guessing here though.
-
Rolls-Royce Optimizes cGAN Training with Databricks MosaicAI
By
–
Rolls-Royce optimized their cGAN training processes with Databricks #MosaicAI. Results included: Drive down TCO Accelerate Time To Model Enhance Model Accuracy and more!
-
Tokenization issues and GPT-4 digit tokenization
By
–
Bytes are still tokens. You’d still have tokenization issues, just new and likely worse ones. Note digits are tokenized specially in GPT-4 as one token per pair. Even in that special case they didn’t go down to single digits.
-
No Important Machine Learning Lessons Expected
By
–
I just don’t think there will be any important lessons for machine learning from it.
-
Speculative RAG Improves Retrieval Augmented Generation Quality
By
–
Speculative RAG splits Retrieval Augmented Generation tasks into two separate steps of drafting followed by verification to provide high-quality answer candidates, and substantially improve the quality and speed of the final output generation. Learn more →
