Zvi's AI #190: Anthropic's code grows exponentially under homegrown "Scion", its internal GitHub replacement
Don't Worry About the Vase (Zvi) · rss · 2026-10-15
Key points from Zvi's weekly roundup:
- Code volume at Anthropic keeps rising exponentially, managed by Scion, an internal GitHub replacement built by a small team in five months
- Google shipped Gemini 4 Argon with a robust safety report; Meta's Muse Wonder has strong benchmarks presented in a "rather sus" way — real-world quality TBD
- OpenAI caught an employee planting a physical bug; it also fired three safety researchers without a good explanation
- A team used ChatGPT to build a full-chain Chrome exploit in six days and claimed a $250k bounty (white hats)
- Pliny let Opus 5.5 run a daily cron job attacking vintage cipher-machine challenges; confirmed solves for Hebern and Kryha problems arrived by day 2
- Also: Epic Games expects little human-written code soon; Microsoft-Decision-1 apparently based on Qwen 3.5-9B; Anthropic Q3 revenue $16.8B; GPUs can be much cheaper to buy than rent; RSI timelines shrink toward 2027
More from Models
- ChatGPT web can actually listen to audio, but the same model can't via API or Codex — Possible_Statement84 · 2026-10-23
- MiniMax M3.1 Flash free for 3 days on 0G, with 1M context and 512K output — MiniMax_AI · 2026-10-23
- TLA+ expert corrects Grok: TLC only exhausts finite state spaces, not all behaviors — tianyin_xu · 2026-10-23
- Open LLM cost calculator shows Arabic tokenization and 6x price gaps across models — Moham-saed · 2026-10-23
- OpenAI blocks manual reset with 13% usage left: 'no reset needed right now' — AccomplishedSugar490 · 2026-10-23
- Meta's 816B Muse Spark Open scores 39, losing to 180B Qwen3.8-Flash-Next at 40 — BTA_Labs · 2026-10-23