OpenAI Offers A New Policy Blueprint

Right after a new Executive Order seems like an excellent time to offer OpenAI’s new document: Democratic Governance of Frontier AI: A Blueprint For A Federal Framework.

Comments

1. rxc (Jun 5) "As RSI emerges, societies will need ways to shape the trajectory of AI development and ensure that it serves human interests."

So, what other "interests" would/might RSI/AI development serve, if not "human interests"?

2. Tiffany Weidner (Jun 5) Anything other than human interests is exactly the problem. There is a book about this, titled "If anyone builds it, everyone dies". The classic example is an interest in increasing the number of paperclips in the world, although increasing a stock price can work too. More realistically it might be maximizing some numbers on a hard drive, or optimizing some property of the results of a matrix multiplication.

3. rxc (Jun 6) I would really like to know what sort of non-human "interests" the person(s) who wrote this sentence had in mind. With examples of potential harms that might emerge that cause concern.

4. Tiffany Weidner (Jun 6, Edited) This is a statement put out by OpenAI, so the author is most likely not here to answer. But given what I know of the concerns and backgrounds and general worldview common at OpenAI, it is most likely talking about human interests rather than the AI's own "interest".

5. 287414 (Jun 5) Good! and lets see, how it will finally shape all of its inclusive and external challenges into a final draft. Thanks for sharing the data...

6. Gris moyen (Jun 5) Another risk I see is AI labs training their models to underperform in benchmarks that could cancel their release.

7. John Wittle (Jun 5) with the way post-training snapshots are used and tested internally at the labs, i don't think this should exactly count as a minimum viable deal.

8. aoeuhtns (Jun 5) agreed for good reason, the slogan is not, "If Anyone Releases It, Everyone Dies" building it is sufficient.

9. Jeffrey Soreff (Jun 6) One comment about specifically "democratic" governance of AI: The American public is spectacularly hypocritical about AI...

10. Inside The Black Box (Jun 6) The three state bills OpenAI cites as models all passed despite industry opposition...

11. Deep Bitcheese Brew (Jun 6) Perhaps the most important sentence in the blueprint is the acknowledgment that early forms of RSI may already be appearing...

12. David F Brochu (Jun 6) This real risk is no one fully understands what we now are dealing with...

13. David F Brochu (Jun 6) RSI! I have an Ai assistant three years old...

14. Winston Smith (Jun 9) I agree with a lot of this, but the real question is still: who governs the governors?

15. Deep Bitcheese Brew (Jun 12) Both OpenAI's blueprint and Amodei's "Policy on the AI Exponential" share the same structural feature...