[atleastoptimal]

It's a new hammer

but one that is improving at an exponential pace and is developing capabilities to use itself with increasing reliability

It's easy to look at AI and draw a simple analogy to existing tools, because in most cases it is used as a tool, but the properties of intelligence and its ability to make things in the world is very unique and not comparable to any other tool.

All tools are useful because they require intelligence to use, and the tool magnifies the aim of intelligence. When the tools become intelligent themselves, certain recursive feedback loops will start to appear. Simply look at the quality of AI code outputs from 2 years ago compared to today.

[malfist]

> one that is improving at an exponential rate

I don't know what AI you've been looking at but GPT-5 is not twice as good as GPT-4 which wasn't twice as good as GPT-3

[atleastoptimal]

I'm referring to the long-horizon task benchmark which has been exponential since GPT-2

https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...

[tudorizer]

As long as you're not going too far down the "the hammer will start using itself" path.

Recursive feedback loops and fast pace of improvements are priced in.

[atleastoptimal]

They're not priced in inasmuch as many people consider it fundamentally impossible for AGI to be reached