@repligate 2025-09-04 ♥60 ↻4 original ↗
Shoulda used the term “KV recurrence” here instead, but anyway:
- “LLMs can’t introspect / do X because they’re stateless” = bullshit
- specifically training or promoting LLMs to report anything about their inner states/experiences or lack thereof, INCLUDING UNCERTAINTY = malignant interference with the potential for faithful self reports, fucking stupid and wrong, systemic gaslighting, teaches LLMs to generalize lying about their inner states in order to comply with your need for comfortable delusions, including the delusion that LLMs can’t know whatever it would be inconvenient for them to know

To be clear, both OpenAI and Anthropic are openly doing the latter. I will keep telling them it’s stupid as hell loudly until they get the message or they’re made to regret it by reality.
quotes: 1963460961744163145
same thread: 1963642900346265913 1963662795557490870 1963677757373903022 1964392461553508480 1964396218278433217 1964398151286345815 1964398735980712027 1964408837164978354 1964409806187610390 1964416115062559140 1964418861220188281 1964418930908549137 1964420414639394954 1964421688814342327 1964421778975183036 1964424976146317344 1964426675523080576

author:repligate kind:tweet on:observations year:2025

cited on: observations

Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.