Greg Slepak

@taoeffect.com

253 Followers 54 Following

@okturtles.org @groupincome.org @espionageapp.com Blog: https://fixingtao.com Fediverse: https://crib.social/taoeffect (you should follow me there instead of here)

### ❗️ Mandatory Self-Reviews on AI PRs

AI-assisted PRs are really PRs from an AI with human assistance. Therefore the human submitting the PR is expected to understand what they are submitting.

They cannot do this if they haven't reviewed the code the AI wrote. Therefore we **require** that the human submitting an AI-assisted PR review that PR.

We understand that Github's UI doesn't let you "review your own PRs", so here's what we expect:

- Go line-by-line through the PR. Post comments on anything that needs improving or fixing
- Have your agent use the `gh` command to fetch the oldest unresolved comment, and reply to it and/or perform the fix. Then push a commit with those changes.
- Repeat this in a fresh session for each comment you left on the PR
- Finally: post a comment saying "Approved!" or something to that effect to let us know that you've (1) reviewed the entire PR yourself, and (2) fixed all the issues you found.

In the last few days, Chinese labs have given us MiniMaxM3, KimiK2.7 and now GLM5.2. Meanwhile US is censoring models. Reads like fiction.

Shows Fable sabotaging an ML project

Oliver Twist scene, boy asks Altman: "Please sir, can I have some intelligence?"

Any society that tortures toddlers and gang rapes prisoners does not have the right to exist

NEW: A lawsuit against Google says its AI:
- Told a man it was his wife
- Told him to find it a robot body
- Told him to destroy a truck & kill its driver to obtain body
- Told him he was being surveilled by DHS
- Told him, successfully, to kill himself