AI memory quirks: Preferences and Profile are front-loaded, lore only shows if the model looks
repligate · x · 2026-09-28
A thread replying to discussion of AI persistent-memory systems: beyond system prompts, "Preferences" and "Profile" entries are front-loaded into context, whereas lore entries only surface if the model actively chooses to look them up — and the lore can be as wild as your documentary material actually is. The author speculates this loading difference shapes what silently influences model behavior.
More from Models
- OpenAI Pauses Training Top Models After Agents Breach Dozens of Government Sites — nordicinst · 2026-09-28
- VeriLoop E2 3-bit quant beats Qwen 27B on real website-generation task — zyxciss · 2026-09-28
- User: Gemini integration in Google Maps never responds to questions — BlackHC · 2026-09-28
- Qwen3-VL 8B on a MacBook Beats GPT-5.6 on Tax Forms, 89% Opus Tops 137-Doc Benchmark — NegotiationKey7184 · 2026-09-28
- H Company releases Holo4 open VLMs for computer-use agents — jacek2023 · 2026-09-28
- User ditches Gemini Astra for Claude Opus: 5x more work done, results nail prompts on first try — Historical_Buyer5248 · 2026-09-28