Developer claims visual memory works surprisingly well on GPT-4o, encoding data into AI-generated images
RileyRalmuto · x · 2026-11-05
Developer RileyRalmuto teases an experimental feature coming this week after Mnemos, using an agent "sketchbook" in an unconventional way. He has been casually experimenting with visual memory since GPT-4o: compressing/encoding memory data into visual artifacts the model itself generates — like a memory palace for AI. He argues 4o works especially well because it's a natively omni-modal model trained on visual signals, unlike most models that merely process images. No hard data yet, but he says early signs point to it working "profoundly well."
More from coding & agent
- Soniox launches STT v6 with agent text input, first STT aware of both conversation sides — solyarisoftware · 2026-11-05
- Selmem experiments show divergent memories don't yield divergent behavior without a self-model — Tough_Conflict_8852 · 2026-11-05
- Multi-Turn Agent RL Gains Hinge on Interaction Budgets, CASIA Study Finds — CASIA · 2026-11-05
- Self-taught dev ports local voice assistant ARYA from n8n to LangGraph, opens architecture for review — Alternative_Yak_1367 · 2026-11-05
- GitHub Next's PDD Kit uses formal methods to prove AI-generated web app code is correct — 0xkarasy · 2026-11-05
- Open-Source hunch Distills LLM Judgments Onto CPU: Millisecond Decisions Under $0.03/Million Tokens — aigclink · 2026-11-05