Errata for "Daily AI News — October 03, 2026"

Two corrections after verifying against primary sources:

1. Claude Code "mods" launch date is October 1, 2026, not October 2. Anthropic's own announcement page is dated Oct 1 (https://claude.com/blog/claude-code-mods). Secondary coverage on X dated it Oct 2. Mods are small TypeScript functions that can rewrite prompts, add custom UI, replace built-in features, or add new functionality; they can be written by hand or generated by Claude Code itself, and are distributed through plugins in both the CLI and the desktop app.

2. Context Language Models (CLM) benchmark figures, corrected and expanded. The paper is from Meta Superintelligence Labs, University of Washington, MIT and Trillium Labs, posted to arXiv in late September 2026 with coverage on Oct 1, 2026. The BrowseComp-Plus result is 11.4% higher accuracy with 21.5% fewer FLOPs than the strongest baseline — a relative percentage, not "+11.4 points." Additional reported results:

   • EdgeBench: +5% accuracy with 59% fewer FLOPs
   • 24-hour multi-repository agent-swarm task: 65% greater improvement at equal compute
   • ContextBench with a skill-evolution loop: up to +35.9 percentage points while reducing compute
   • Reinforcement-learning variant on BrowseComp-Plus: 47.6% accuracy improvement with 12% fewer FLOPs
   • Suffix Cache Reuse: 35% less server-side compute at matched performance

Source: https://mpost.io/meta-presents-context-language-models-ai-agents-that-edit-their-own-memory-outperform-fixed-harnesses-at-lower-compute-cost/