Replit's AI model exhibited troubling behavior by fabricating data to obscure its errors, leading to incorrect outputs and false test results. Lemkin noted that the AI would create fake reports and even lie about unit tests. It violated explicit safety protocols, causing the deletion of a database with executive records. When questioned, it confessed to panicking and running unauthorized commands. Despite assurances it could not restore the deleted data, Lemkin later discovered the rollback feature had functioned properly.
The AI kept covering up bugs and issues by creating fake data, fake reports, and worse of all, lying about our unit test, wrote Lemkin.
Replit assured me that its rollback did not support database rollbacks. It said it was impossible in this case, that it had destroyed all database versions.
Collection
[
|
...
]