percent

I believe I touched on that in my last sentence, but I can elaborate:

That will probably happen – neural networks are approximation algorithms. It's a question of how often that happens.

What percentage of the calls get misclassified? And what's the threshold percentage that would be needed for the triage bot to be a net positive?

It sounds like they have an idea of these numbers based on data collected from the non-emergency line, so it's not like they're just blindly jumping into this.

EDIT: I just realized that I did not actually answer your question of "what happens"...

I imagine the caller would just interrupt the AI's answer (e.g. "No not that," "HELP," "Give me a human," "FUCK!" etc.)? That seems like the natural thing to do.

To be clear: I don't know anyone at Carbyne or OPCD. I can only offer speculation.

Holy shit, I'm actually surprised how bad the rest of the comments are... And the volume of them!

I'll try to TL;DR it for the lemmings, with formatting that is (hopefully) easy to understand for even the most rotten of brains.

TL;DR

The problem

• Too many people are all calling 911 about the same emergency.

I'll use this example scenario below: People keep driving past a burning car on a busy road, and many of them call 911. (This will continue to happen until an emergency responder arrives.)

Before implementing this tech

• 911 operators are all busy answering calls that are all reporting the same car fire
• Long 911 hold time for someone with an emergency unrelated to the car fire

After implementing this tech

Bot: "Are you calling about the car fire on Seventh Street?"

• If caller answers "yes":
  • AI bot tells them that responders have already been dispatched
• If caller answers anything other than "yes":
  • Transfer to the next available human dispatcher
  • Greatly reduced hold time thanks to automated triage

If the critical failure point is accurately classifying "yes" or "not yes," even the dumbest^1 models could handle that – and I doubt they use the dumbest models for 911 triage.

Even if it's not 100% perfect every time, this still sounds like a net positive.

I ask this out of genuine, burning curiosity, and mean no rudeness by this...

Did you read the article that the OP linked to?

I don't mean to single you out with this question, I'm just confused after scrolling this far through sooo many similar comments. It almost feels as if I somehow ended up reading a different article from everyone else.

It sounds like that's a better alternative than what they had before though, no?

Side note: either I'm wildly misunderstanding something in the article, or a BIG majority of commenters here didn't actually read it. I'd expect some amount of the latter, but it really seems disproportionate here.

How many? And how does that number compare to how many people died due to call volume surges before implementing this triage tech?

So, Lemmy really seems to love these two things:

• Getting angry at things
• Having their biases confirmed

If a headline feeds the hunger, why risk ruining that validation by reading the article?

It seems like they're doing this because they are prioritizating the public safety aspect of the job, no?

Advertisers might find a way to sneak ads into prompt injections or something

They definitely can be useful as a part of a "productivity machine," for some people. But yeah, they're not the whole solution. An LLM alone is much less useful without harnesses and tools.

I suppose one of the difficulties in trying to build a one-size-fits-all thing is that everyone has a different set of use cases and workflows needed for productivity. Each person's own "productivity machine" might be useless for other people.

Most people (maybe even most tech-oriented people) just aren't AI-fluent enough to be able to build their own agentic workflows with enough fidelity to actually yield a net-positive effect in productivity. It's not a user-friendly process for the masses (yet).

Yep, CAD and house building are still human jobs, for now.

Maybe

...this assumes that no one will ever connect LLMs to machinery, or generate code to run machinery (gcode, for example)?