[Aerroon]

Even as a teenager the singularity didn't make sense to me. Why would AI capabilities suddenly improve by an astronomical amount? It's still going to be running on hardware and software that we (humans) design. Even if AI reached parity with human brains it would still need to improve efficiency by orders of magnitude to reach parity with human society's output. And only then could it start to outpace the improvements that humans are making.

Even if the AI singularity is possible, the acceleration would take a long time. Possibly a lifetime.

[Izkata]

The singularity is the tipping point where the AI can improve itself faster than humans can comprehend, so the software at least wouldn't be human-designed anymore.

It requires a level of self-driven intent and access to their own code LLMs aren't capable of on their own. They might be one component of such a future system though, we just don't know yet. Even then it would probably take quite a while to reach that tipping point. GP is just wrong, talking as if we'd already reached it.

[danny_codes]

We won't know how close we are to takeoff until takeoff. My main critique of AI is that takeoff is dangerous, and we're basically running towards the cliff as fast as we can. Will you grow wings before you fall to your death? That's basically the bet we're making, if you believe in takeoff.

Silver lining is that if it goes bad we probably won't live long (as a species) to say "I told you so"

[mike_hearn]

Probably not though. AI is already designing its own chips, optimizing its own GPU kernels, suggesting the next experiments to run and writing its own harnesses... recursive self improvement is here already. It's just a question of speed, time and how far it goes.