One AI to rule them all?
retained text
The title of this thread even makes me uncomfortable, but hear me out. I'm in the camp that unregulated models (local models) pose the biggest threat in the near-term. A local model can have guardrails removed and enable a crazy person to make Covid but 10x more deadly, or at least, could hypothetically enable this more in the coming decade. I don't foresee someone being able to use ChatGPT to do that in the near future, but local models are risky, and they only lag 1-2 years behind frontier flagship. There's also no stopping local models with current policy and humans. Cat is out of the bag, as compute gets cheaper and more abundant, how do you stop XYZ people from creating powerful unregulated models in the year 2030? I don't see how. This generates the conversation. The only tool with more power than humans, in theory, is ASI once built. A super cop to rule them all and actively monitor emerging AI tools, and remain two steps ahead in terms of security and censorship. I'm also fully aware of how risky that sounds - literally handing off power and policing to a single model, that's basically the premise for all Terminator movies/skynet, but I also don't know of any other way to prevent local models which will pose a huge threat. I do sometimes worry that AI could be the great filter and reason we don't see populated galaxies when we look up. I'm mostly an accelerationist, but, can't ignore this thought...
document fetched 2026-10-24 00:31 · canonical https://reddit.com/r/singularity/comments/1xds9d9/one_ai_to_rule_them_all