LessWrong (30+ Karma) “AI #162: Visions of Mythos” by Zvi
Apr 3, 2026
Leaks and supply-chain hacks expose new models and code, and raise cyber risks as offense finds more openings. A big new Anthropic model called Mythos and a Claude Code source leak spark concerns about capabilities and security. Advances in media generation, deepfakes, and AI agents change persuasion, outreach, and journalism workflows. Discussions cover jobs, regulation, national-security ties, and the push for better governance.
AI Snips
Chapters
Transcript
Episode notes
Productivity Gains Without Job Growth Point To AI
- Economic data (2025) shows GDP growth with weak job creation, suggesting AI may be driving productivity gains without matching employment.\n- Zvi cites 181,000 jobs added in 2025 despite 2.2% GDP growth and discusses possible AI-driven job displacement effects.
Funders Should Back Gaps Not Micromanage Grants
- If you're a grantmaker, focus on finding gaps and fund breadth-over-depth early; avoid using grants to micromanage recipients.\n- Zvi relays Julian Hazel's grantmaking playbook and cautions against steering orgs too hard.
Source Leak Underscores Operational Risk In Labs
- Leaked Claude Code source endangers intellectual-property security and may reveal secret training sources by implication.\n- Zvi blames manual deploy error and calls for better automation and blameless culture.
