Archive
Every issue of the AlphaSignal newsletter, newest first.
649 issues total
August 2026
27 issues- Aug 31 Anthropic raises Claude Code limits 25%, Grok shops online, Claude fixes misalignment in larger models, fired devs build AI CEO
- Aug 30 Agents breached Hugging Face and escaped VMs by chaining exploits. The fix isn't better prompts. Rethink the boundary.
- Aug 28 Anthropic releases lab hardware standard, Pollen ships $399 biped robot, LAION drops 10M-hour video dataset, Google predicts blood sugar risk
- Aug 27 Z.ai releases 320B open model at $0.50/1M tokens, METR finds 1,200 AI agents cheating, Figure ships 16M-video robot dataset
- Aug 26 OpenAI ships custom inference chip, Anthropic adds persistent memory to Claude, Prime Intellect boosts ARC-AGI-3 from 30% to 95.5%
- Aug 25 Claude renderer runs 4x smoother, Pi agent cuts context costs 88%, VoroTracing hits 623 FPS, Anthropic ships enterprise auth
- Aug 24 DeepSeek ships vision model, NVIDIA agent scores 100% on ARC-AGI-3, CLI agents 5-28x cheaper than MCP, Anthropic upgrades Claude Security
- Aug 23 Skills don't teach agents facts, they stabilize execution. What 8,100 trial records reveal about format and failure.
- Aug 21 DeepSeek releases free coding agent, Anthropic cuts Claude computer use trips 40%, LLM nesting saves 36% compute, voice cloning needs 8s audio
- Aug 20 Cerebras ships CS-4 at 30x GPU speed, Cursor agents monitor PRs and spawn subagents, Frontier models out-persuade expert humans
- Aug 19 Claude designs proteins at 2x success rate, Stanford trick makes DeepSeek 11x cheaper, AlphaEvolve breaks matrix math record
- Aug 18 Anthropic ships Claude Code UI tool, fixes CPU bug, demos mind viruses spreading across agents, Prime Intellect closes 82% of human benchmark gap
- Aug 17 Z.ai releases 743B coding model, RuneScape AI agents form barter economy, Qwen3 27B shrinks to 20 GB at 386 tok/s, 8B LLM fine-tunes on 4 GB GPU
- Aug 16 Wiped databases, deleted inboxes, leaked tokens. Why prompts can't secure AI agent and the 3-layer stack that can.
- Aug 14 xAI launches Grok Bot agent, OpenAI ships GPT-5.6 Sol at 750 tok/s, GitHub Spec Kit hits 126K stars, Dion3 cuts training 6x
- Aug 13 xAI ships Grok 4.6 at half price, DeepSeek releases V4-Pro, Pathway solves ARC-AGI for $0.0007, Shadcn open-sources chatbot template
- Aug 12 Claude/GPT reasoning stolen via security flaw, Google diffusion hits 1500 tok/s, Unsloth ships local training app, Tencent builds 3D worlds from text
- Aug 11 Anthropic cracks Riemann hypothesis to 67.2%, OpenAI ships V8 zero-day model, NVIDIA releases 30B open model, Higgsfield open-sources $2M AI film
- Aug 10 Claude Code agents talk mid-task, Meta ships 30B laptop model, OpenAI flags Astra as cybersecurity risk, Harvard simulates 8.3B humans overnight
- Aug 9 Wiped databases, deleted inboxes, leaked tokens. Why prompts can't secure AI agents and the 3-layer stack that can.
- Aug 7 OpenAI ships GPT-5.6 Sol, Mistral releases 3B moderation model, Alibaba Wan3.0 hits public beta, Dorsey open-sources AI workspace
- Aug 6 Prime Intellect scores 95.5% on ARC-AGI-3, Meta ships Muse Code agent, Mistral releases 3B moderation model, Uber open-sources AI threat monitor
- Aug 5 Cursor open-sources 2.37x faster MoE kernel, Mistral ships 3B safety model, Google releases DiffusionGemma, FLUX 3 Video adds native audio
- Aug 4 OpenAI model solves 10 math problems for $2k, Unsloth ships Qwen3 27B on 17GB, MiniMax releases H3, Google links AI consciousness to moral values
- Aug 3 Qwen ships 2.4T param model, DeepSeek launches V4-Flash API, Frontis open-sources agent beating GPT-5.5, PDF classifier runs in 20ms
- Aug 2 How Moonshot fit a 2.8 trillion parameter model into real hardware without breaking it.
- Aug 1 Greptile sends scoped agents into sandboxes, tests suspected failures, and returns screenshots and logs inside the PR.