<< Go back to HN Digests
Warning : This document is still a draft

2026 W15 - 04-12 - Mythos




First time here ? Check the introduction here

What you Missed this Week

This week, Anthropic announced a new model for cybersecurity: Mythos (Link, Model Card). Restricted to only a few companies, for vulnerability discovery. At the same time, user complaining about quota melting too quickly and other bugs and claude code.

At the end of this same week, Aisle said that they were able to identify the vulnerabilities with very small models (5B Link)

Articles

Model Training / Tunning

Model improvement or other low level optimization

  • 390 pt [Link / HN] Muse Spark: Scaling towards personal superintelligence
  • 230 pt [Link / HN] Show HN: Gemma 4 Multimodal Fine-Tuner for Apple Silicon

Coding Agents

Claude new functionalities, tips and tricks dedicated to a specific agent, …

Claude

  • 1528 pt [Link / HN] Project Glasswing: Securing critical software for the AI era
  • 1343 pt [Link / HN] Issue: Claude Code is unusable for complex engineering tasks with Feb updates
  • 1210 pt [Link / HN] Small models also found the vulnerabilities that Mythos found
  • 839 pt [Link / HN] System Card: Claude Mythos Preview [pdf]
  • 479 pt [Link / HN] Pro Max 5x quota exhausted in 1.5 hours despite moderate usage
  • 448 pt [Link / HN] Claude mixes up who said what
  • 424 pt [Link / HN] I’ve been waiting over a month for Anthropic to respond to my billing issue
  • 371 pt [Link / HN] Anthropic downgraded cache TTL on March 6th
  • 321 pt [Link / HN] Assessing Claude Mythos Preview’s cybersecurity capabilities

  • 222 pt [Link / HN] Claude Code login fails with OAuth timeout on Windows

  • 167 pt [Link / HN] Claude Managed Agents

The rest

  • 207 pt [Link / HN] ChatGPT Pro now starts at $100/month
  • 148 pt [Link / HN] Tell HN: OpenAI silently removed Study Mode from ChatGPT

Personal Blogs / Opinions

Investigation

Trend investigation, hacking, data analysis, …

  • 905 pt [Link / HN] Show HN: I built a tiny LLM to demystify how language models work
  • 469 pt [Link / HN] Exploiting the most prominent AI agent benchmarks
  • 324 pt [Link / HN] MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU
  • 200 pt [Link / HN] Research-Driven Agents: When an agent reads before it codes
  • 145 pt [Link / HN] AI helps add 10k more photos to OldNYC

Tool for AI / Made with AI / showHN

Any Github / ShowHN for new tools / product

  • 466 pt [Link / HN] Show HN: Ghost Pepper – Local hold-to-talk speech-to-text for macOS
  • 340 pt [Link / HN] Reallocating $100/Month Claude Code Spend to Zed and OpenRouter
  • 322 pt [Link / HN] Launch HN: Freestyle – Sandboxes for Coding Agents
  • 199 pt [Link / HN] Instant 1.0, a backend for AI-coded apps
  • 161 pt [Link / HN] Show HN: CSS Studio. Design by hand, code by agent
  • 155 pt [Link / HN] Show HN: Gemma Gem – AI model embedded in a browser – no API keys, no cloud
  • 128 pt [Link / HN] Show HN: Hippo, biologically inspired memory for AI agents
  • 108 pt [Link / HN] Show HN: Marimo pair – Reactive Python notebooks as environments for agents

Optimists

  • 313 pt [Link / HN] AI assistance when contributing to the Linux kernel
  • 150 pt [Link / HN] Productive Procrastination

Pessimists

  • 615 pt [Link / HN] The cult of vibe coding is dogfooding run amok
  • 604 pt [Link / HN] ML promises to be profoundly weird
  • 312 pt [Link / HN] AI Will Be Met with Violence, and Nothing Good Will Come of It
  • 132 pt [Link / HN] The AI Great Leap Forward

  • 129 pt [Link / HN] The Future of Everything Is Lies, I Guess: Part 3 – Culture

  • 109 pt [Link / HN] OpenClaw’s memory is unreliable, and you don’t know when it will break

Other

  • 409 pt [Link / HN] I still prefer MCP over skills
  • 269 pt [Link / HN] The Vercel plugin on Claude Code wants to read your prompts
  • 265 pt [Link / HN] The future of everything is lies, I guess – Part 5: Annoyances
  • 263 pt [Link / HN] Taste in the age of AI and LLMs
  • 231 pt [Link / HN] AI may be making us think and write more alike
  • 183 pt [Link / HN] Your File System Is Already A Graph Database

Officials

Politics / Strategy

Merge, acquisition

  • 356 pt [Link / HN] Sam Altman’s response to Molotov cocktail incident
  • 275 pt [Link / HN] Cirrus Labs to join OpenAI

News

Anything from “official” journals

  • 2151 pt [Link / HN] Sam Altman may control our future – can he be trusted?
  • 430 pt [Link / HN] OpenAI backs Illinois bill that would limit when AI labs can be held liable
  • 286 pt [Link / HN] Anthropic expands partnership with Google and Broadcom for next-gen compute
  • 244 pt [Link / HN] AI singer now occupies eleven spots on iTunes singles chart
  • 227 pt [Link / HN] Google open-sources experimental agent orchestration testbed Scion
  • 227 pt [Link / HN] Molotov cocktail is hurled at home of Sam Altman
  • 163 pt [Link / HN] Copilot is ‘for entertainment purposes only’, per Microsoft’s terms of use

Unrelated

Not related to AI, but worth reading

  • 684 pt [Link / HN] I run multiple $10K MRR companies on a $20/month tech stack
  • 375 pt [Link / HN] S3 Files
  • 309 pt [Link / HN] The Importance of Being Idle

  • 119 pt [Link / HN] Will I ever own a zettaflop?

  • 115 pt [Link / HN] Moving fast in hardware: lessons from lab to $100M ARR

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