Hacker News new | ask | show | jobs
by wizzwizz4 543 days ago
LLM-based AI is not needed, or even useful. We know how to make voice interfaces that work, and work well: have done since the 80s. It's just expensive; and it's an expense that nobody in the industry is willing to pay, therefore nobody needs to do it in order to differentiate their product.
1 comments

What you're missing is that AI solves the expense problem. As the OS vendor you already have an overview and easy access to all interfaces that you expose and it's straightforward to feed that into an integrated AI agent. Add a bit of glue code here and there and a simple implementation is nearly free. Of course, the real value lies in ironing out all the edge cases, but compared to doing all of that manually, it should still be orders of magnitude cheaper.
It's not, because "ironing out all the edge-cases" is orders of magnitude more expensive than just designing a system without edge-cases in the first place. What's cheap is getting away with not bothering: but then you end up with a tech demo, rather than a usable product.