Eugene Yan

how do you evaluate if an RQ-VAE is trained well? i see papers mention codebook use (these have >90% use) and reconstruction loss. do y'all also look at validation loss, quantization loss, final residual, or something else? p.s. pink line with the lowest loss uses rotation trick…

This day four years ago we adopted Latte. She came earlier than expected and we had to build a makeshift crate lol

what a world we live in! I just took a Jupyter notebook that implements an llm-evaluator, provided it as context to a coding assistant, and then ask it to write an evaluator class with specified inputs, outputs, etc. initially, it was verbose with too many class methods, but with…

Claude Code is getting a brand new feature: custom subagents. Type `/agents` to get started.

Working with world’s most respected engineers & researchers on the world’s most challenging problems is tech’s a house closer to the beach in Amagansett. We don’t need made up goals like the finance industry because we actually work on real problems.

listening to @vibhuuuus and @eugeneyan jam on kimi-k2 paper. highest alpha free ai nerd-snipe content on the internet. @latentspacepod paper club

tools for learning and building • zotero: papers, annotations, citations • obsidian: notes, writing, connecting the dots • claude: thinking partner, improve understanding • claude code: implementation, code review (images from work on semantic IDs and residual-quantized…

criminally underrated talk from @devanshtandon_ Gemini powers YouTube's Large Recommender Model by **tokenizing every video on youtube** (SemanticID) - a vocabulary several OoMs larger than English, CONTINUOUSLY PRETRAINED every day. can reason across titles/descriptions, throws…