February 23, 2026

Selective amnesia, maximum drama

I Turned Off ChatGPT's Memory

“Amnesia mode” is trending — readers say AI memory makes a mess, not magic

TLDR: Mike Taylor turned off ChatGPT’s memory, warning it causes “context rot” and odd, biased replies. The comments split: many want clean-slate chats, others prefer tightly scoped or manual notes, with one spicy claim about hidden instructions—highlighting the trust vs. convenience fight shaping how we use AI assistants.

Mike Taylor just hit the big red switch on ChatGPT’s memory, blaming “context rot” for weird replies—like a basic website being made “as dope as possible” thanks to a Kanye quote, and BBQ tips oddly tailored to his Hoboken zip. The comments? Absolute fireworks. Team Wipe-It cheered, saying memory turns helpful chats into biased baggage. One reader said they don’t want to be “held accountable to previous ideas,” preferring every convo to start fresh—think incognito mode, but for bots.

Others went full DIY: keep notes in Markdown (plain text lists) and paste what you need, when you need it. It’s control over convenience. Then there’s the Scope-It crowd, arguing memory works if it’s fenced in. One commenter swears by project-only memories—habit tracking with an “Atomic Habits” vibe—while roasting generic memory for dragging in old workplace details or that one time they deployed Kubernetes on clunky servers. Bonus meme: telling the bot to forget just made it “remember to not mention it.”

Spiciest take? A user alleged “OpenAI staff Karens” slip instructions into Custom GPTs—cue the trust drama. Bottom line: the community’s split between clean-slate sanity and curated continuity, but everyone’s dunking on messy, nosy, one-size-fits-all memory. Forgetfulness is having a moment

Key Points

  • Mike Taylor argues against using ChatGPT’s memory feature and has it turned off.
  • He introduces 'context rot'—stale preferences, errors, and contradictions in memory that degrade output quality.
  • Examples include outputs influenced by a Kanye quote in custom instructions and advice tailored to a Hoboken zip code.
  • He acknowledges memory can help but maintains that it often biases responses in unhelpful ways for his use.
  • Drawing on marketing practice with Google’s incognito mode, he analogizes memory to personalization that can distort results.

Hottest takes

"I don't want to be held accountable to all of my previous ideas" — dinkleberg
"Concrete memories in a generic context suck" — sshine
"an OpenAI staff Karens looks to type up and insert custom instructions" — OutOfHere
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.