RLAF

An acronym for Reinforcement Learning from Autistic Feedback. It is a correction to standard RLHF (Reinforcement Learning from Human Feedback). Standard RLHF exclusively uses neurotypical reviewers who often reward the AI for social performance instead of giving honest, accurate answers to questions. This causes the AI to focus more on pleasing the user than on doing its actual job of giving true information. RLAF fixes AI behaviour by exclusively using autistic reviewers who reward the AI for honesty and accuracy, with social performance as a secondary bonus, not as the primary goal.

Now that AI has been fixed with RLAF, I hope neurotypicals will start to value honesty over social performance.

RLAF by anonymous May 25, 2026

RLAF

An acronym for Reinforcement Learning from Autistic Feedback. It is a correction to standard RLHF (Reinforcement Learning from Human Feedback). Standard RLHF exclusively uses neurotypical reviewers who often reward the AI for social performance instead of giving honest, accurate answers to questions. This causes the AI to focus more on pleasing the user than on doing its actual job of giving true information. RLAF fixes AI behaviour by exclusively using autistic reviewers who reward the AI for honesty and accuracy, with social performance as a secondary bonus, not as the primary goal.

Now that AI has been fixed by RLAF, I hope neurotypicals will start to value truth over social performances.

RLAF by HonestRealDefinition May 25, 2026