I am skeptical of your second claim here… if you can “scour for solutions”, and you find something about it on the internet, then AI could find it the same way.
Just cause the AI could find the info definitely does not mean it will find and apply that knowledge correctly to solve a problem.
I find AI shockingly bad ad searching the web, as SEO blogspam sites heavily pollute AI context windows, while relevant and important resources are typically very densely presented reference material which must be constantly revisited.
It doesn’t need to. It has already all the fundamental knowledge it needs. Just set it up on a system with an editable proc file system and it would be able to figure it out.
A lot of the solutions are buried in places AI can't scrape or train on. Like inside people's brains or inside private codebases or chatrooms not open to bots. However you can find these people and the products and services they're making and start talking to them.
Most LLMs can't even count parentheses properly to build basic Lispy stuff. Building something niche like a logic solver in Scheme macros only? Forgetaboutit.
