Mastodawn

RE: https://furry.engineer/@soatok/115651773214269698

Since May is halfway over, I think my timeline projection was wrong, but the AI companies are certainly playing with fire.

They're currently doing what might be called "enshittification" if they were any other SaaS: lower usage limits, pricing restructuring to capture more value for themselves, etc.

This isn't the sort of thing that will kill off the use of LLM technology or even generative AI as a whole. So-called "local models" will likely persist the AI bubble pop.

But what won't survive is the service model where AI users' data is funneled into a third-party service provider and processed by an LLM.

Show thread

Soatok Dreamseeker May 18

It's hard to gauge exactly when this might happen.

The AI companies are certainly positioning themselves for a massive bail-out, hoping the Trump administration will hand them trillions of dollars because they're "too big to fail" if enough CEOs force their employees to adopt this tech.

(Which logically means people who resist it are a threat to their scheme.)

Show thread

Rycochet May 18

@soatok If the bailout isn't bad enough, just imagine the partisan strings that will be attatched. Look forward to the grokification of all AI models, folks.

Show thread

tekhedd May 19

@Rycochet @soatok Oh yes, you don't think anyone in authority wants us downloading and running our own unbiased, uncensored models, do you?

For AI to be a success for the ruling classes and corps, a *usable* AI has to have hardware requirements well out of reach of any end user. This puts them in control of the models and lets them tax every use, keeping small players from competing *or* having badthink.

Show thread

tekhedd

@Rycochet @soatok TBH I do really believe they're trying to apply "monopoly via superscale" to AI, but I also don't think this will work. This does not work (for long) with technology, because engineers.