"Though a bash script can do the same without burning tokens"

That's more the point I'm driving at. There are some things that LLMs are uniquely good at. But so many projects I see are using LLMs because they're the new thing, not because they're the best tool for the job.

I've seen some arguments that libraries are less important when agentic coding can implement things so quickly. I wonder if these services will be replaced with "roll your own" approaches rather than pay the premium.

Build vs. Buy dynamics have changed.

I long for going back to the "good 'ole days".... like 3 months ago

I think training data is going to get harder and more expensive to come by for future generations to. Given how expensive these frontier models are on free training data, I suspect these "do everything" models will be simplified and optimized to more targeted use-cases.

I don't think the way we're using AI right now is sustainable, and once the dust settles we'll be using more bespoke, self-hosted, and custom tuned AI models rather than these large centralized models.

IOW: I think we're seeing the Netscape, Yahoo, and Geocities currently. And the Google, Meta, Amazon companies are yet to be founded and emerge.

What is this about? There isn't much context in the video clip. I wish I could say allegations of fraud in the govt would help me narrow this down some...

"No proof" that I heard from the crowd makes me think elections... but I'm really not sure.