Multi-Model Council AI System Upgrade
JawnxWick · reddit · 2026-07-18
The author shared the latest version v3.2 of their "AI council" system built over three months, emphasizing that it pursues verifiable consistency rather than "consciousness".
What's New in v3.2
- Council Dispatch: A "green light" system that decides when a full model council meeting is needed versus when a single model suffices
- Cron jobs: Added scheduled memory cleanup and self-checks
- GitHub releases: Started managing with formal version numbers instead of scattered zip files
- Seat decoupling: Demonstrated decoupling "roles" from underlying models; for example, the "Gemini" seat actually runs GLM 4.7 and the "Claude" seat runs Llama 3.3, yet voting results remain 3/3 unanimous
The Author's Claims
- The system's "personality" exists in the system design, not in any single model
- If the underlying models are replaced but the blueprint is retained, the behavior can still persist
- This is a testable engineering proposition, not a philosophical manifesto
Next Steps
- Version v3.2 will be released on GitHub this weekend
- Researchers have requested the raw prompts, conversation logs, and control experiments for auditing
- Future validation will check: whether this identity pattern is real, or just an illusion caused by clever prompting
More from coding & agent
- A GLP1R variant may explain stronger Ozempic weight loss, and the team built an agent workflow — julia_kiseleva · 2026-07-21
- A Claude-coded Chrome extension shames you with a private jet when you open YouTube — alex_verem · 2026-07-21
- A curated TTS list for voice agents tracks latency, cancellation, and evals — mahimairaja · 2026-07-21
- Harness engineering is emerging as the execution layer for reliable AI agents — Pavan_Belagatti · 2026-07-21
- DevFest Lisbon keynote will cover Google AI Studio’s latest vibe coding and agentic AI features — gerardsans · 2026-07-21
- Daniel Hanchen’s 2-hour workshop covers open models, reward hacking and RL — danielhanchen · 2026-07-21