This is a special post for quick takes by Elliot Callender. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
One could argue that restricting consumer access reduces transparency/awareness and transparency/awareness regarding capability might have been useful in mobilizing the public toward a pause. If the public experiences agentic, highly capable AI firsthand, you may increase the number of people worried about job replacement, RSI leading to an intelligence explosion, offensive capabilities outpacing defensive capabilities, and misalignment more broadly.
Even if we have whistleblowers in the companies warning us, without model access, a large portion of the public will brush these reports off as hype.
1
1Weaker arguments I thought up:
- Reducing consumer access reduces the odds of a meaningful warning shot. This logic seems cursed to me, and I don't want to live in a world where we actively try to make things worse to make things better.
- Pause advocates will not be able to use frontier models to do pause advocacy work. I think this one is a cost, but probably offset by (1) many accelerationists also not having access and (2) the big labs always retaining the advantage here anyways with their best model being leveraged to lobby for their preferred policies.
1Reducing consumer access reduces the odds of a meaningful warning shot. This logic seems cursed to me, and I don't want to live in a world where we actively try to make things worse to make things better.
I think that, in isolation, this wouldn't significantly affect the change of a warning shot, but would make it more likely the public noticed it. The effect is probably pretty weak though.
(Updated thread to clarify that outsider advocates should have access, I'm curious about consumers as a separate thing.)
Okay, that makes sense. I had a cached prior that {autoregressive LLMs can't continual-learn
My own take on restricting consumer access as good is that--if the trend continues--it will send strong signals that the tech is unusable at scale due to safety concerns. This raises the probability that the AI bubble pops. I think there are futures where the bubble popping now would be bad, but I mostly think it would be good. Everything will depend on how the public would interpret the pop, with one group loudly proclaiming that this AI-stuff was a potemkin village built of hype and another group saying that AI is the real-deal and safety issues caused the pop.
Edit: Meh, on second thought, the safety concerns might dent the market case for AI but it would significantly strengthen the national security case for AI, so the government would most likely step in and fund the major labs regardless if the bubble really popped on safety concerns.