9+ hour, 29+ min ago   (76+ words) AI inference chip company competing directly with Nvidia. Builds the LPU, a Language Processing Unit based on a Tensor Streaming Processor architecture with 230MB on-chip SRAM and no HBM, using static compile-time scheduling to eliminate GPU non-determinism. Delivers Llama 3 70B at roughly…...

2+ day, 21+ hour ago   (189+ words) An unreleased Claude research model pushed a 40-year-old mathematical lower bound from 41.6% to 67.2%, using a multi-agent swarm running inside Claude Code The Riemann zeta function is a mathematical object that encodes the distribution of prime numbers. It has infinitely many…...

3+ day, 3+ hour ago   (655+ words) We ran DeepSeek V4 Flash through 54 private debugging attempts, then compared its reliability, cost, latency, token use, and patch behavior with nine other frontier models. DeepSeek V4 Flash has been making a lot of noise in the AI community since it launched…...

1+ week, 6+ day ago   (515+ words) AlphaSignal - Google Research introduced the Science One Framework, an experimental autonomous research prototype that eliminates hallucinated citations and unreproducible results. - The framework is built on Chain-of-Evidence (CoE), which requires every claim in an AI-generated paper to be traceable to a…...

2+ week, 1+ day ago   (151+ words) Goodfire's Silico platform used Direct Logit Attribution to cut an RL mode collapse from 94% to 5% in one afternoon, without retraining Reinforcement learning runs collapse all the time. A model latches onto a shortcut, starts parroting the same phrase in 94% of…...

2+ week, 1+ day ago   (285+ words) AlphaSignal - Liquid AI releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, open-weight bidirectional encoders with 8,192-token context. - LFM2.5-Encoder-230M is 3.7x faster than ModernBERT-base on CPU at 8k tokens (28s vs 90s per forward pass). - LFM2.5-Encoder-350M ranks 4th of 14 models on GLUE/SuperGLUE/multilingual benchmarks, behind only three larger models....

2+ week, 1+ day ago   (33+ words) Rehman Rind AlphaSignal Try modifying the filters AI moves faster than any field in history. AlphaSignal tracks every paper, repo, model, and release in real time—organized, searchable, and tuned to your work....

2+ week, 2+ day ago   (329+ words) AlphaSignal Epoch's FrontierMath Benchmark Cracks a 40-Year-Old Math Problem Using Fable 5 Epoch AI has announced that AI has found a presentation for the absolute Galois group of the field of 2-adic numbers, a problem that has resisted mathematicians since the…...

3+ week, 1+ day ago   (54+ words) AI safety nonprofit focused on pre-deployment risk evaluation of frontier models. Conducts autonomous-capability assessments for Anthropic, OpenAI, Google, and Meta, testing whether models can self-replicate, conduct cyberattacks, or resist shutdown. Maintains TaskFamily, a benchmark of multi-hour agentic tasks, and the…...

4+ week, 1+ day ago   (237+ words) Today, vLLM is introducing vLLM prefill paired with TileRT decode, integrated through vLLM V1's public connector interface and shipping with TileRT 0.1.5. The headline is simple: you can now route latency-sensitive traffic to a purpose-built decode engine while keeping everything else,…...