<< Go back to HN Digests

2026 W3 - 01-18




First time here ? Check the introduction here

Articles

Model Training

  • 704 pt Link / HN TimeCapsuleLLM: LLM trained only on data from 1800-1875
  • 760 pt [Link / HN] AI generated music barred from Bandcamp
  • 585 pt [Link / HN] Pocket TTS: A high quality TTS that gives your CPU a voice
  • 388 pt [Link / HN] Flux 2 Klein pure C inference
  • 222 pt [Link / HN] Raspberry Pi’s New AI Hat Adds 8GB of RAM for Local LLMs
  • 189 pt [Link / HN] Furiosa: 3.5x efficiency over H100s
  • 174 pt [Link / HN] Show HN: OSS AI agent that indexes and searches the Epstein files
  • 138 pt [Link / HN] vLLM large scale serving: DeepSeek 2.2k tok/s/h200 with wide-ep
  • 107 pt [Link / HN] Show HN: Sparrow-1 – Audio-native model for human-level turn-taking without ASR
  • 100 pt [Link / HN] Reproducing DeepSeek’s MHC: When Residual Connections Explode
  • 58 pt [Link / HN] Batmobile: 10-20x Faster CUDA Kernels for Equivariant Graph Neural Networks

Coding Assistant

  • 1225 pt [Link / HN] Cowork: Claude Code for the rest of your work
  • 823 pt [Link / HN] Claude Cowork exfiltrates files
  • 293 pt [Link / HN] Claude is good at assembling blocks, but still falls apart at creating them
  • 264 pt [Link / HN] We put Claude Code in Rollercoaster Tycoon
  • 259 pt [Link / HN] Scaling long-running autonomous coding
  • 209 pt [Link / HN] First impressions of Claude Cowork
  • 203 pt [Link / HN] Show HN: OpenWork – An open-source alternative to Claude Cowork
  • 156 pt [Link / HN] The recurring dream of replacing developers
  • 143 pt [Link / HN] Provide agents with automated feedback
  • 138 pt [Link / HN] Letting Claude play text adventures
  • 99 pt [Link / HN] Bubblewrap: A nimble way to prevent agents from accessing your .env files
  • 105 pt [Link / HN] Using proxies to hide secrets from Claude Code
  • 101 pt [Link / HN] Show HN: Webctl – Browser automation for agents based on CLI instead of MCP
  • 95 pt [Link / HN] Show HN: Yolobox – Run AI coding agents with full sudo without nuking home dir
  • 96 pt [Link / HN] Show HN: Agent-of-empires: OpenCode and Claude Code session manager
  • 71 pt [Link / HN] Show HN: Gambit, an open-source agent harness for building reliable AI agents

Personal Blogs / Opinions

  • 364 pt [Link / HN] We can’t have nice things because of AI scrapers
  • 340 pt [Link / HN] Anthropic made a mistake in cutting off third-party clients
  • 225 pt [Link / HN] A letter to those who fired tech writers because of AI
  • 185 pt [Link / HN] Data is the only moat
  • 183 pt [Link / HN] The insecure evangelism of LLM maximalists
  • 138 pt [Link / HN] Let’s be honest, Generative AI isn’t going all that well
  • 70 pt [Link / HN] LLMs are a 400-year-long confidence trick
  • 63 pt [Link / HN] AI Destroys Institutions
  • 54 pt [Link / HN] Why we don’t use AI
  • 52 pt [Link / HN] Two Concepts of Intelligence

Politics / Strategy

  • 961 pt [Link / HN] Apple picks Gemini to power Siri
  • 268 pt [Link / HN] Anthropic invests $1.5M in the Python Software Foundation
  • 155 pt [Link / HN] Mozilla’s open source AI strategy

Other

  • 333 pt [Link / HN] LLM Structured Outputs Handbook
  • 195 pt [Link / HN] FLUX.2 [Klein]: Towards Interactive Visual Intelligence
  • 69 pt [Link / HN] Confer – End to end encrypted AI chat

News

  • 296 pt [Link / HN] Signal leaders warn agentic AI is an insecure, unreliable surveillance risk
  • 181 pt [Link / HN] Games Workshop bans staff from using AI
  • 175 pt [Link / HN] Google removes AI health summaries after investigation finds dangerous flaws
  • 138 pt [Link / HN] Tldraw pauses external contributions due to AI slop
  • 89 pt [Link / HN] Superhuman AI Exfiltrates Emails
  • 88 pt [Link / HN] Instagram AI Influencers Are Defaming Celebrities with Sex Scandals

  • 67 pt [Link / HN] OpenAI to test ads in ChatGPT as it burns through billions

  • 51 pt [Link / HN] The Risks of AI in Schools Outweigh the Benefits, Report Says

>> You can subscribe to my mailing list here for a monthly update. <<