Even just a year ago, the idea that you could gain any value for AI Alignment from studying/playing with LLMs was considered not serious by most people in or adjacent to AI Safety.
But I think it's now clear that this empirical hands-on approach does pay off.
But I think it's now clear that this empirical hands-on approach does pay off.
Well, I think that a well-funded, principled, and grounded AI Alignment research program that studies DMT entities is a natural next step.
They can in fact be human-level or higher in their intelligence. Faster gestalts, hyperdimensional medium of thought, no ego defenses.
They can in fact be human-level or higher in their intelligence. Faster gestalts, hyperdimensional medium of thought, no ego defenses.
And in many ways (as argued in the latest Qualia Computing post) also quite similar to the simulacra LLMs spin up for the sake of prediction.
More so, DMT entities don't scale (you can't "copy" them a million times or have them infiltrate the internet!). So they might be safer.
More so, DMT entities don't scale (you can't "copy" them a million times or have them infiltrate the internet!). So they might be safer.
Just a thought...
Planting seeds, you know?
Planting seeds, you know?
• • •
Missing some Tweet in this thread? You can try to
force a refresh
