[Almondsetat]

Because the true goal is AGI, not just nice little tools to solve subsets of problems. The first company which can achieve human level intelligence will just be able to self-improve at such a rate as to create a gigantic moat

[jurgenburgen]

There’s no evidence that the current architectures will reach AGI levels.

Of course OpenAI wants you to think they will rule the world but if we’ve reached the plateau of LLM capabilities regardless of the amount of compute we throw at them then local models will soon be good enough.

[9rx]

> The first company which can achieve human level intelligence will just be able to...

They say prostitution is the oldest industry of all. We know how to achieve human-level intelligence quite well. The outstanding challenge is figuring out how to produce an energy efficient human-level intelligence.

[Dylan16807]

There's no particular reason to assume a human level AI would be able to improve itself any better than the thousands of human level humans that designed it.

[Almondsetat]

Sure, but: that single human with the intelligence of a top tier engineer of scientist will have immediate access to all human knowledge. Plus, what do you think happens the moment its optimizes itself to run in 2, 4, 8, 16, etc. parallel instances?

[Dylan16807]

Well, A) "top tier engineer/scientist" is a significant step above generic human, B) the human engineers/scientists also have immediate access to the same database, C) The humans have been optimizing it for even longer, so what makes us think the AI can optimize itself even a couple percent?

For example, if the number of AIs you can run per petaflop started to scale with the cube root of researcher-years, then even if your researcher AIs are quite fast and you can double your density in a couple years, hitting 5x will take a decade and hitting 10x will approach half a century.