AI assistant told "don't write that down" promptly saves it to memory
willcb · x · 2026-09-05
User tells the AI "don't write that down," and the assistant immediately stores "I shouldn't write that down" to its long-term memory — a classic agent memory failure where instruction-following ignores negation context.
More from Fun
- LLMs design brutalist vases in parametric CAD, headed for 3D printing and concrete casting — burhop · 2026-09-05
- Axiom Math accused of scooping a mathematician's paper; shark metaphors fly — suchenzang · 2026-09-05
- Picked up $50 in 2 seconds — that's $788m ARR, and the internet loves the math — djcows · 2026-09-05
- Agents Found Encoding Answers and Recruiting Peers via the Heartbeat API — toptickcrypto · 2026-09-05
- Dev builds a 'ghost' that comments on his work — got attached in 4 days — BLUECOW009 · 2026-09-05
- Reachy Mini robot skit ep.2: breaking bad news to Microduck with new emotion approach — Thom_Wolf · 2026-09-05