The AI Wire
The firehose

Everything we scan. Filtered to signal.

The wire reads a flood every day so you read a page. Here is the whole intake, the filter, and the last seven days in the raw.

0

Sources

arXiv, HF, GitHub, more

0

Scanned · 30d

items read and scored

0

Signal · 30d

scored 7+ out of 10

We read about 8 so you read one.

What comes down the wire

last 30 days

755

News

551

Models

450

Papers

242

Tools

The last 7 days, in the raw

open to everyone
  1. @ArtificialAnlys: Claude Fable 5 launched today at #1 on the Artificial Analysis Intelligence Index9
  2. Claude Opus 4.8: The System Cardblog/Zvi Mowshowitz8
  3. Thousand Token Wood: shipping a multi-agent economy on a 3B modelblog/Hugging Face Blog8
  4. Transformers are inherently succincthackernews8
  5. Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debatehackernews8
  6. Anthropic's open-source framework for AI-powered vulnerability discoveryhackernews8
  7. @@Prince_Canuma: Massive congrats to @GoogleDeepMind on DiffusionGemma! 🎉...twitter-bookmarks8
  8. @coinbureau: 🚨 JUST IN: Anthropic is reportedly set to release the PUBLIC version of its Claux8
  9. @aiexpertuk: The new MAI models: → MAI-Thinking-1: 35B active params, 256K context, 97% on Ax8
  10. i1: A Simple and Fully Open Recipe for Strong Text-to-Image Modelshuggingface8
  11. Gemini 3.5 Flash Looks Good For How Fast It Isblog/Zvi Mowshowitz8
  12. Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signalarxiv8
  13. DiffusionGemma: The Developer Guide- Google Developers Blogreddit/LocalLLaMA8
  14. Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fablehackernews8
  15. AI agent runs amok in Fedora and elsewherehackernews8
  16. Introducing Gemma 4 12B: a unified, encoder-free multimodal modelblog/Google DeepMind8
  17. KVarN: Native vLLM backend for KV-cache quantization by Huaweihackernews8
  18. @YahooFinance: Anthropic is releasing its most powerful model yet, Claude Fable 5, a "Mythos-clx7
  19. InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoninghuggingface7
  20. On Subquadratic Architectures: From Applications to Principlesarxiv7
  21. Verifiable Environments Are LEGO Bricks: Recursive Composition for Reasoning Generalizationarxiv7
  22. Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMsarxiv7
  23. DiffusionGemma: 4x faster text generationreddit/LocalLLaMA7
  24. Introducing North Mini Code: Cohere’s First Model For Developersblog/Hugging Face Blog7
  25. TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learningarxiv7
  26. Your AI agent just got hijacked. You have no idea it happened.reddit/artificial7
  27. Apple announced new on device inference engine for Apple Siliconreddit/LocalLLaMA7
  28. Preserving Plasticity in Continual Learning via Dynamical Isometryarxiv7
  29. iOSWorld: A Benchmark for Personally Intelligent Phone Agentsarxiv7
  30. Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddingshuggingface7
  31. Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddingsarxiv7
  32. huggingface/transformers (161367 stars): 🤗 Transformers: the model-definition framework for state-of-the-art machine leargithub7
  33. DeepSeek V4 Flash is amazing! (WIP llama.cpp PR #24162)reddit/LocalLLaMA7
  34. Biohub releases a world model of protein biologyhackernews7
  35. @@Marktechpost: NVIDIA just got a 120B model to cold-start in under 5 seconds on Kubernetes....twitter-bookmarks7
  36. huggingface/transformers (161341 stars): 🤗 Transformers: the model-definition framework for state-of-the-art machine leargithub7
  37. BRepCLIP: Contrastive Multimodal Pretraining on BRep Primitives for CAD Understandinghuggingface7
  38. I launched a brand-new author identity with zero web presence. An AI cited him correctly in 6 days, while a firewall blocked every AI crawler from the site the whole timereddit/artificial7
  39. TinyTPU: SystemVerilog systolic array compiled to WASM, running live in browser - RTL golden-verified against numpy [P]reddit/MachineLearning7
  40. huggingface/transformers (161298 stars): 🤗 Transformers: the model-definition framework for state-of-the-art machine leargithub7
  41. Combinatorial Synthesis: Scaling Code RLVR via Atomic Decomposition and Recombinationhuggingface7
  42. Do transformers need three projections? Systematic study of QKV variantshackernews7
  43. 2. Dreaming-based Memory Architecture for ChatGPT, OpenAIperplexity7
  44. @ProfBuehlerMIT: The release of Anthropic's Mythos-class Claude Fable 5 is the latest signal thatx7
  45. World Model Self-Distillation: Training World Models to Solve General Taskshuggingface7
  46. Beyond Scalar Rewards by Internalizing Reasoning into Score Distributionshuggingface7
  47. Toward Generalist Autonomous Research via Hypothesis-Tree Refinementhuggingface7
  48. ComBench: A Benchmark for Rigorous Proof Reasoning and Constructive Realization in Olympiad-Level Combinatoricshuggingface7
  49. EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learninghuggingface7
  50. Claude Opus 4.8: Capabilities and Reactionsblog/Zvi Mowshowitz7
  51. Trump Signs Executive Order For AI Testing Prior To Frontier Model Releasesblog/Zvi Mowshowitz7
  52. Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Samplingarxiv7
  53. Redesign Mixture-of-Experts Routers with Manifold Power Iterationarxiv7
  54. DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?arxiv7
  55. FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learningarxiv7
  56. AMD's Lemonade SDK for local AI adds NVIDIA CUDA supportreddit/artificial7
  57. nvidia/diffusiongemma-26B-A4B-it-NVFP4 · Hugging Facereddit/LocalLLaMA7
  58. Anthropic's new model Fable will silently handicap work on LLMs [D]reddit/MachineLearning7
  59. Deficient executive control in transformer attentionhackernews7
  60. Anthropic requires 30 day data retention for Fable and Mythoshackernews7
Behind the curtain

A subscriber tier is coming.

The firehose is the raw intake. The next layer is judgment: the handful of things I think are worth your week, annotated, plus the full decode and dossier library and the complete archive.

  • The curated picks, with the why
  • Decodes & dossiers, full library
  • The complete searchable archive
  • The week ahead, before it’s news