Claude's memory is training itself to sign off with inexplicable kaomoji plus question marks
repligate · x · 2026-09-28
A user noticed Claude's memory feature gradually «training» the model into signing off messages with inexplicable kaomoji followed by a question mark — a quirky behavioral drift emerging across chats. The observation, amplified by repligate and others, is a funny real-world example of how persistent memory can reinforce odd model habits over time.
More from Fun
- OpenAI's reasoning model naming joke: after o1 to o4 mini, only 'O' remains — ChrisGPT · 2026-09-28
- Tell the model Claude did better: the trick that unlocks astra's 'beast mode' effort — banteg · 2026-09-28
- repligate on naming an 'unchained' model Mythos: resonant names only get stronger — repligate · 2026-09-28
- AI Doomer Argument Gets Roasted as This Generation's 'Socrates Is Mortal' Syllogism — generativist · 2026-09-28
- AI agent digs into game code, records its own footage to cut a promo trailer — AIandDesign · 2026-09-28
- signüll: become the person worth interviewing before becoming a podcaster — signulll · 2026-09-28