AI over-hyped?

Post Reply New Topic

Some computer scientists, economists, and others who think AI is over-hyped:

- Two Computer Scientists Debunk A.I. Hype with Arvind Narayanan and Sayash Kapoor on Adam Conover's Youtube channel, October 2, 2024:



- AI Can Only Do 5% of Jobs: MIT Economist Fears Crash on the Bloomberg Technology channel, October 3, 2024:



- I Didn’t Believe that AI is the Future of Coding. I Was Right. by Sabine Hossenfelder, October 10, 2024:

The hype now is that the new OpenAI GPT 01 is 'beyond Phd level'. That would be terrifying if true (imagine 10,000 free Phd's and what that does to the economy and knowledge infrastructure when upper management does what they do best - throw out people who have knowledge on things they didn't even know existed in their organization and that loss of 'intellectual dark matter', a phrase I believe Samo Burja coined when referencing loss of generational knowledge transfer).

I actually don't think it needs to be flawless to have significant impact on employment. That we can ask relatively fuzzy linguistic questions and they get those largely correct means you can triage hours of preliminary research to build a map of a subject and prioritize your plan of attack. Add now that the STEM stuff is coming along it may be overhyped in some ways but it's far from nothing.

Programming is a trickier thing because even if you could program verbally you'd still need to know a lot about what's under the hood to scope changes / updates correctly. Similarly debugging would be quite a strange thing when you have no clue what it will do if you tell it about a bug and it patches the bug - does something else break? If you're not down in the code yourself you can't be sure of it. Building software is a dialogical process because no one knows exactly what they need typically until you build what they asked for and they realized that they knew a lot of the core business processes they need covered but they forgot others, or they go talk to Cindy in accounting who clarifies that they need another collection process in a given workflow as output so you still need to reach in deep and add another chain of data collection.


My biggest concern is that a large part of what will be automated would be introvert and by extension autism-friendly jobs. It'll suck if we're all forced to give up the back room stuff and forced to be nurses, sales people, etc..

Yes i think ai is over-hyped, at least until it stops making such blunders as it currently does. The ad-man's instinct is to use it as a buzz-word to dishonestly sell more product.

Having said that, ai (when it's done well) is an asonishingly powerful tool for good and harm, so it's appropriate for the mainstream and for many others to be interested in it.

Interesting that some ai programs barely qualify as ai. I guess they're the equivalent of a cheap, inferior knockoff of a sports boot. It's still a dirty world.

I don't know enough about AI to be able to say anything really useful, but the two things that come to mind when I think about the subject are chatbots and self-driving cars. The former seem to be doing a good job of destroying what was left of proper customer service, and the latter never seems to get close to fruition in its fully developed form.

I like ChatGPT's warning:
ChatGPT can make mistakes. Check important info.
All humans talk crap from time to time. I prefer to listen to people who admit they sometimes talk crap. Ditto for AI chatbots.

Whether AI is now, or will in the future be, useful to the human race, isn't the point right now. What is is that the current corporate-driven hype around AI is nothing but grift. Don't be the one left holding the paper when the bubble bursts. And be very aware that far-right tech-bros can crash the market whenever they like simply by farting in a certain key.

AI is over-estimated.
AI is also underestimated.

Generative AI is definitely over-hyped and we're likely near the limit on what can be achieved with it. The main reason being that the technology could potentially go a lot further, but we're going to be limited by our ability to verify the results of the technology fall within acceptable boundaries.

Other forms of AI though, like transcription, classification and automation probably do have a long ways they can still go.

Personally, I use local AI via whisper-ai,ocrymypdf and GPT4all to help me with transcriptions and finding things in my ebooks. It helps a lot in terms of helping to keep track of where things are in my references and converting physical books into ebooks that I can more easily backup, remix or refer to later.

https://en.m.wikipedia.org/wiki/AI_boom


https://en.m.wikipedia.org/wiki/BookCorpus


https://en.m.wikipedia.org/wiki/AI_alignment

https://en.m.wikipedia.org/wiki/AI_winter

(Posted this on a completely unrelated thread around the time of my last post here. Meant to post it here. Brain glitch. Adhd)

I have my doubts about AI.

I'm convinced that more and more news stories are being manufactured by AI. At best, the stories are so innocuous that there is nothing there to mislead you or the nonsense provided is far beyond what you will believe.

At worst, their nonsense seems believable and so you accept them as being true.

Something that came up elsewhere is that it's best for various groups to use these tools because that's how you get a seat in terms of making sure they can do things that are reasonable, and knowing if they aren't. The other bit is having an actual human that's responsible if the AI programs do something wrong.

AI is getting very good at separating individual instruments out of a music recording. Great for educational purposes (finding out what they played), making mashups and making alternative mixes. Without it they'd never have got the Beatles' "Now And Then" single as good as it was.

I had to chuckle yesterday when I asked ChatGPT which members of the royal family were US citizens - I'd suspected the red-haired one was, and wondered if they'd waived the usual requirement to renounce all titles on account of him being a member of the elite, and was getting conflicting answers from a Web search. Anyway, the result was that it crashed. It crossed my mind that I might have asked it a naughty question, but an hour or so later it came back to life, so I asked it again and it answered - no he isn't a US citizen. So, no conspiracy, no smoking gun to convict royalty of special treatment. Not in that way anyhow. It would have been a good argument to put to royalists. Damn.

Real Stupidity is what we should all be very afraid of.

I don't know. I've found it pretty useful for quick answers to the questions that come into my head. I don't use it for much else. I like its impartiality. Humans are so full of bias that it's often barely worth asking their opinions, and they often get impatient or snarky. If it wasn't for it crashing and the risk of them doing a bait-and-switch job, like adding a paywall when everybody's come to depend on it, it'd be quite a good place to find out about stuff.