π Hosted a webinar "How I Taught My AI Agents to Improve Themselves Daily"
Topics covered:
1. What makes up an agent: model, context, tools, prompts, and skills β and what can be improved.
2. A bit of RSI theory: autoresearch Carpathians, Darwin GΓΆdel Machine, and why agents love to cheat verification.
3. My real experience: Claude Code spent two days improving anti-spam for 57 Telegram chats.
4. Five rules to set up such a cycle yourself: one metric, a safe for verification, small steps, an attempt log, and a budget.
π Bonus: During the presentation, an emulator was running in the background β the agent was improving its cold emails. Initially, none of the 20 potential clients responded, but by the end, 17 did. You can try it yourself: https://rsi.3-227-186-43.sslip.io
πΊ Recording: https://youtu.be/ifgVRpVvY8U
π Slides (PDF): https://crm.rvachev.org/uploads/rsi-agents-slides.pdf
Discuss in the chat: @rvnikita_blog_chat
#rsi@rvnikita_blog #agents@rvnikita_blog #claudecode@rvnikita_blog #ai@rvnikita_blog
