Story Archive

Stories accumulated over time. Reachable from timeline spikes too.

181 stories

10-31 Validated K8s runtime for NVIDIA Dynamo ・ NVIDIA 2 sources
10-30 Easier edge inference and Dynamo runtime ・ Inference 6 sources
10-28 Sleep method aids LLM continual learning ・ Machine Intelligence 2 sources
10-28 NVIDIA tunes Nemotron to local rules ・ NVIDIA 3 sources
10-27 AWS, Google open PyTorch for own chips ・ Google 5 sources
10-24 NVIDIA widens TensorRT tuning scope ・ NVIDIA 2 sources
10-23 OlmoEarth v1.3 adds global embeddings ・ Embeddings 4 sources
10-22 Agent Skills go to Agentic AI Foundation ・ AI Agents 3 sources
10-22 Microsoft solves the KLS conjecture ・ Microsoft 3 sources
10-21 Agent dev tooling expands ・ AI Agents 11 sources
10-21 OpenAI models coming to Google Cloud ・ Google 2 sources
10-21 NVIDIA extends AI factory tooling ・ NVIDIA 8 sources
10-21 Cohere ships Rerank 5 for retrieval ・ Software Engineering 4 sources
10-20 ChatGPT Ads expands to 26 new countries ・ GPT 5 sources
10-20 Mistral releases Voxtral 3 ・ Mistral 2 sources
10-20 Rethinking LLM confidence and fairness ・ Software Engineering 8 sources
10-17 DeepSeek releases V4.1-Flash ・ DeepSeek 2 sources
10-16 Meta Muse Spark scales on NVIDIA GB300 ・ Meta 4 sources
10-16 Granite Switch 4.1 unifies three modes ・ Reinforcement Learning from Human Feedback (RLHF) 4 sources
10-15 Google Deep Review parses tech docs ・ Machine Intelligence 2 sources
10-14 On-device agents and coding agent teams ・ AI Agents 11 sources
10-14 Anthropic opens Singapore office, MOU ・ Anthropic 5 sources
10-14 Apple runs distributed LLM on Mac Studio ・ Fine-tuning 4 sources
10-14 Google launches Gemini 4 Argon ・ Gemini 3 sources
10-14 DeepMind, Singapore launch anti-scam AI ・ Google 5 sources
10-13 Nadella: SI models are internal risk ・ Microsoft 3 sources
10-13 Idle GPUs and state reuse speed decoding ・ NVIDIA 2 sources
10-09 Embed/Rerank: shared vs dedicated ・ Embeddings 4 sources
10-08 Anthropic expands science, cyber efforts ・ Anthropic 5 sources
10-08 NVIDIA cuOpt scales to 100M variables ・ NVIDIA 4 sources