“On ‘Responsible Scaling Policies’ (RSPs)” by Zvi

This post was originally intended to come out directly after the UK AI Safety Summit, to give the topic its own deserved focus. One thing led to another, and I am only doubling back to it now.

Responsible Deployment Policies

At the AI Safety Summit, all the major Western players were asked: What are your company policies on how to keep us safe? What are your responsible deployment policies (RDPs)? Except that they call them Responsible Scaling Policies (RSPs) instead.

I deliberately say deployment rather than scaling. No one has shown what I would consider close to a responsible scaling policy in terms of what models they are willing to scale and train.

Anthropic at least does however seem to have something approaching a future responsible deployment policy, in terms of how to give people access to a model if we assume it is safe for [...]

---

Outline:

(00:17) Responsible Deployment Policies

(03:15) How the UK Graded the Responses

(04:22) Anthropic's Policies

(05:27) The Risks

(10:42) The Promise of a Pause

(13:58) ASL-3 Definitions and Commitments

(18:16) Approaching Thresholds

(24:38) ASL-4

(27:26) Underspecification

(29:06) Takeaways from Anthropic's RSP

(35:30) Others React

(38:30) A Failure to Communicate

(39:47) OpenAI Policies

(41:56) DeepMind Policies

(45:53) Amazon, Inflection and Meta

(47:53) Some Additional Relative Rankings

(48:57) Important Clarification from Dario Amodei

(55:07) Strategic Thoughts on Such Policies

(01:05:37) Conclusion

---

First published:
December 5th, 2023

Source:
https://www.lesswrong.com/posts/yRJNCDp7LHyHGkANz/on-responsible-scaling-policies-rsps

---

Narrated by TYPE III AUDIO.