The A.I. Thread

[Russic]

Originally Posted by TorqueDog

Looks like it was running on good ol' GPT-4o which is probably not great for this sort of thing. I'll have to give it another try using the other models, I didn't even think to run them through o3 or 4.1.

EDIT: Yup, WAY better on 4.1. It actually did what it was supposed to, with no hallucinations.

That's great to hear! 4o is SO good at creative writing, quick low-stakes questions, or creative brainstorming. Anything mission critical though needs to be kept far away from it.

What's great is OpenAI just made o3 way more efficient, so even us lowly plus users get about 100/week. I'm pretty confident if you started up your own business, o3 could handle many of the tricky, pitfall-rich "getting started" tasks for you (marketing plans, business roadmaps, competitor research, etc.).

Granted, bolting this stuff into existing infrastructure is not always a good time. But starting something from scratch? Lordy, what a different story in many cases.

As for the chess thing, it's mostly all fun and games. I worry a lot though that people are given a false comfort when they should be re-evaluating what they do. Not total freak out bash-each-other-in-the-skulls-and-dig-a-bunker, but maybe ask the question "what if my job as a call center support agent didn't exist in 2 years?"

So I worry that stories about hilarious AI failures, while entertaining, might be doing many of us a concerning disservice.

[Fuzz]

but maybe ask the question "what if my job as a call center support agent didn't exist in 2 years?"

Ya, that's a question I hope policy experts and governments are grappling with, because it is absolutely going to eliminate a lot of low level jobs people who are already struggling need.


I know it's impossible because of our global economy, but a tax on AI agents that save companies millions of dollars would be reasonable, otherwise they just get a massive efficiency boost and society(taxpayers) deal with the fallout.


The reality is many people just can't do jobs that require intelligence over a certain point. You can't just say train and educate. Some people are just dumb.

[DoubleF]

Originally Posted by Wormius

I am not sure how exactly Apple is doing it, but I have tons of photos on my phone and I was able to search for one really easily instead of trying to scroll through thumbnails or try to narrow down where the photo was.

I meant if AI might start producing a bunch of digital trash at unprecedented levels which further amplifies issues with data storage/transmission/energy usage.

[Itse]

Originally Posted by Fuzz

The quoted Bluesky post is essentially making the point of using the right tool for the job, and while LLM's can be good at a lot of things, it's not good at all things. Which is probably a good message for clueless execs looknig at deploying these tools because they hear they can do everything.

The main selling point of LLM "AI" kind of is their ability to do "anything". Their convenience comes primarily from it being a multitool so you don't have to learn or create a million separate tools, let alone know what other tools are out there.