Sublime

An inspiration engine for ideas

AllPeopleCollectionsArticlesAudioBooksFilesHighlightsImagesLinksNotesTextTweetsVideosSocial

New paper from Meta shows them beating everything Llama-based on the leaderboard with a new approach - that is unsupervised! There are also be deeper applications to this, and a new class of models that might be coming up Also funny... See more

Hrishi

x.com

Today we’re announcing Meta LLM Compiler, a family of models built on Meta Code Llama with additional code optimization and compiler capabilities. These models can emulate the compiler, predict optimal passes for code size, and disassemble code. They can be fine-tuned for new optimizations and compiler... See more

AI at Meta

x.com

so many interesting ideas/gems in this project! i think mechanically it solves and points to a lot of directions: prompt as genetic code and hence evolving material, visual outputs as phenotype for CLIP to measure novelty. not to mention the experiments of using research into creativity strategies and... See more

Kat ⊷ the Poet Engineer x.com

StyleGAN2 projector modifications, with CLIP, addl. losses, and other cool things... this time as a gist, maybe eventually I'll have enough stuff for a repo. Feel free to play with it! cc: @l4rz @dvsch @advadnoun @erinbeess @theshawwn https://t.co/1polsbiqMb https://t.co/6PYsA9gLCQ

Peter Baylies x.com

contrary to popular believe, integrating sonnet 4.5 into your app is not a one line model string change! some things we learned at cognition from using 4.5 early in devin (long running agent) and windsurf https://t.co/s3lBs2Ofk3

andrew gao

x.com

CM3: A Causal Masked Multimodal Model of the Internet abs: https://t.co/NkmRCGTcbB sota in zero-shot summarization, entity linking, and entity disambiguation. generate images unconditionally, conditioned on text (like DALL-E) and do captioning all in a zero-shot setting... See more

x.com

Introducing `llama-405b-to-8b` ✍️ Get the quality of Llama 3.1 405B, at a fraction of the cost and latency. Give one example of your task, and 405B will teach 8B (~30x cheaper!!) how to do the task perfectly. And it's open-source: https://t.co/H5590RiFhc... See more

Matt Shumer x.com

realising the beast that PPO is https://t.co/vNvSdhlXte

Chinmay Kak

x.com

Introducing DeepConf: Deep Think with Confidence 🚀 First method to achieve 99.9% on AIME 2025 with open-source models! Using GPT-OSS-120B even without tools, we reached this almost-perfect accuracy while saving up to 85% generated tokens. It also delivers many strong advantages for parallel... See more

Jiawei Zhao x.com