10 Comments
User's avatar
Robin Thacker's avatar

You're asking the right questions. Keep posting.

Everest's avatar

Particularly interested in:

9. Marginal risk as BS

11. Evals as BS

10. Post-scarcity as BS

24. Response to AI as normal technology

25. Response to Karnofsky on Anthropic's RSP 3.0

8. What a good future might look like

1. What is happening right now? What is everyone doing and why?

19. The societal immune system

22. Tool AI wants to become agent AI

20. My experiences with COVID

14. Why don't people seem worried about out-of-distribution generalization?

5. On rogue AI vs "scheming"

13. AI won't stay limited to internal deployment

4. Why is AI risk such a hard problem?

William's avatar

I'd love to see more on 3. Are you going to get into why they should prove safety, using their insider knowledge, rather than expect concerned outsiders to with an information disadvantage to argue the case for risk?

I wonder how 14 connects with the paper arguing that beyond a certain complexity, LLMs are incapable of carrying out computational and agentic tasks or verifying their accuracy?

John Wittle's avatar

i wouldn't mind some engagement with janus's (perhaps justified) crashout recently against kyle fish et al

i've been wanting to see what an orthodox safetyist thinks of human p(doom [deserved])

i know it's not on your list, but as long as you're eliciting honest feedback from your readers...

(that being said, please don't hook me up to an emotion-vector honest-feedback-elicitation mind probe)

Piotr Zaborszczyk's avatar

1,2,4,5,6,7,8,9,10 (I suspect I might disagree with you here), 11, 14, 15, 16, 17, 18, 19, 21

Bradford Saad's avatar

I’d be particularly interested in 4, 6, 7, 8, 12, 15, 17, 20, and the response posts you mention.

Kathleen v's avatar

1,4,6,8,10 to start

torchbearercommunity's avatar

We would be interested in a post about how you structure a meeting with a lawmaker. We have a training program to teach the DIP (Direct Institutional Plan) from ControlAI to anyone who wants to learn how to talk to lawmakers. We want to improve this using real-world experience.

Also, any insights about how to build a university course on AI safety. We might have an opportunity to do this.

Kevin's avatar

1, 6, 7, 10, 16, 18, 22

Your thoughts on "AI as normal technology" and Anton Leicht would also be good.

rif a saurous's avatar

These all sound good! I am most interested in 1, 3, 4, 6, 8, 9, 10, 11, 12, 21, 22, 23.

I'd also be quite interested in a discussion of the word "pause", which I think has become a worryingly overloaded word that can refer to "a few months until we solve alignment" or "an arbitrarily long period of time until we solve extremely hard problems that we don't know how to formulate yet."