Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new epoch in…
.abbel-fig { display: block; text-align: center; margin: 2.4em 0; line-height: 1.4; max-width: 100%; } .abbel-fig img { display: block; margin: 0.65em auto 0; height: auto; max-width:…
A toy dynamical model of whether the AI workforce that builds future AI ends up cooperative or uncooperative: where the basin boundary lies, what current evidence…
... government of the people, by the people, for the people ... — Abraham Lincoln, Gettysburg Address (1863) The cost of AI is dropping rapidly. GPT-4-class…
Congratulations to the Berkeley Artificial Intelligence Research (BAIR) Lab class of 2026! This year, BAIR celebrates another remarkable group of Ph.D. graduates whose curiosity, creativity, and…
ParallelKernelBench tests whether LLMs can write fast multi-GPU CUDA kernels across 87 real workloads. The best model solves under a third, but a few generated kernels…
We generated 12 landing pages with Kimi K2.7 Code and Claude Fable 5. Kimi cost 94% less and scored within a few points on every page.…
Together AI has earned ISO 27001:2022 certification, validating our commitment to enterprise-grade security for production AI workloads.
How Together served MiniMax-M3 efficiently with KV-block-major sparse attention, paged MSA decode, optimized index scoring, and a Rust-based multimodal gateway.
.apr-fig { text-align: center; margin: 1.35em 0; line-height: 1.4; } .apr-fig--wide img { display: inline-block; width: 100%; max-width: 100%; height: auto; vertical-align: middle; } .apr-fig--wide-0-8 {…
.grasp-results-table table { font-size: 0.875rem; line-height: 1.35; width: 100%; } .grasp-results-table th, .grasp-results-table td { padding: 0.35rem 0.5rem; } /* Consistent whitespace between major sections (this…
Using importance sampling with fine-tuned donor prefills to predict reward hacking emergence during training
Understanding the behavior of complex machine learning systems, particularly Large Language Models (LLMs), is a critical challenge in modern artificial intelligence. Interpretability research aims to make…
An encoder (optical system) maps objects to noiseless images, which noise corrupts into measurements. Our information estimator uses only these noisy measurements and a noise model…
In this post, I’ll introduce a reinforcement learning (RL) algorithm based on an “alternative” paradigm: divide and conquer. Unlike traditional methods, this algorithm is not based…
Interim report on ongoing work on reward hacking
What exactly does word2vec learn, and how? Answering this question amounts to understanding representation learning in a minimal yet interesting language modeling task. Despite the fact…
Announcing Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
.modal { display: none; position: fixed; z-index: 9999; padding-top: 50px; left: 0; top: 0; width: 100%; height: 100%; overflow: auto; background-color: rgba(0,0,0,0.9); } .modal-content { margin:…
Research update on on applying local volume measurement to downstream tasks
In this post, we will study inductive biases of the parameter-function map of random neural networks using star domain volume estimates. This builds on the ideas…
Announcing the Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text
Using Product Key Memories to encode sparse coder features
Recent advances in Large Language Models (LLMs) enable exciting LLM-integrated applications. However, as LLMs have improved, so have the attacks against them. Prompt injection attack is…
Open-weight model releases. The most important shifts in AI through 2025–26 happened here — Llama-3.x, Qwen3, DeepSeek-V3, GLM-4.5 and Kimi-K2 closed the gap with frontier closed models on coding and reasoning benchmarks while being downloadable. We track the Hugging Face blog, EleutherAI, BAIR, Allen Institute for AI (Ai2 / OLMo), and GitHub release feeds of the major model repositories.
52 stories indexed in this category. See also all models and the full source catalogue.