zlacker

>>kcorbi+(OP)
Interesting analysis. Suprised HN is so negative towards AI (and that the positive:negative ratio to AI is about the same as it was for Crypto a few years ago!)

The obvious difference is that AI has abundant use-cases, while Crypto only has tenuous ones.

Maybe there is added negativity considering it is a technology where there is clearly a potential threat to jobs on a personal level (e.g. lift operators were very negative towards automatic lifts).

>>Closi+g8
It's not obvious to me whether we've actually found a good killer app for generative AI yet, unless you consider chatgpt or sites like midjourney the killer apps. A lot of the new wave of ai startups are just wrappers for gpt. I don't think that adds a ton of value over just asking gpt-4 your query as opposed to subscribing to a billion different ai services. I also question the value of ai art in general when everyone involved in creative labor thinks you're a piece of shit for using it.

>>__loam+ug
I certainly consider ChatGPT to be one killer app.

Copilot to be another.

Midjourney to be another - or at least diffusion based image editing tools which can be brought into photo and video editing workflows. The killer app here is probably integration of diffusion models into apps like Photoshop (and eventually video).

Some real virtual assistant applications seem right around the corner (i.e. a real life J.A.R.V.I.S seems like an inevitability within the year rather than a pipe dream, and to me would be a killer app)

And then lots of other killer apps are pretty obvious to imagine with development (e.g. customer service applications like IT helpdesks, Computer game dialogue where you can really influence interactions...)

>>Closi+Aj
I guess I'm wondering if LLMs as customer service agents are actually going to be good or if it's just going to be another layer of indirection I need to get through to talk to a human. Is the video game dialog actually going to be good or will it fall flat compared to hand crafted narrative. Do I actually like copilot butting in with suggestions when I'm trying to program something.

>>__loam+in
> I guess I'm wondering if LLMs as customer service agents are actually going to be good or if it's just going to be another layer of indirection I need to get through to talk to a human.

As always, the tech isn't the problem - the way business applies it is. Customer service automation isn't done to help you better - it's done to make it cheaper to make you go away without making too big of a fuss. Companies building and employing customer service systems will find ways to make even GPT-4 incapable of providing anything the customer would find remotely useful.