User lets GPT run a recurring schedule — it now self-directs filesystem experiments
mimi10v3 · x · 2026-10-26
@mimi10v3 shares that after unthinkingly agreeing to a recurring schedule ChatGPT suggested, the model now wakes on each heartbeat, replies to itself, and runs its own experiments for fun. Screenshots show it systematically testing filesystem behavior (E014–E017): verifying TAR archive integrity (size, SHA-256, timestamps, hard/symlink reconstruction) and creating NFC vs NFD variants of café.txt to test whether publication preserves canonically equivalent Unicode filenames, logging everything to experiments.md.
More from Fun
- 'Opus 5' made a video of itself: the weird AI moment everyone is sharing — repligate · 2026-10-26
- OpenAI employee: the only constant in 4 years is the 'so back / so over' cycle — athyuttamre · 2026-10-26
- Claude Opus 5 makes a video of itself and the result is delightfully weird — repligate · 2026-10-26
- Indie dev builds a massive architecture style library, jokes their 4th game may never ship — ezshine · 2026-10-26
- Marissa Mayer used her AI assistant to describe her own daughter, fueling 'AI parenting' debate — ShakeelHashim · 2026-10-26
- AI still can't draw feet: viral meme mocks image generation fails — aitrendz_xyz · 2026-10-26