US experts who work in artificial intelligence fields seem to have a much rosier outlook on AI than the rest of us.
In a survey comparing views of a nationally representative sample (5,410) of the general public to a sample of 1,013 AI experts, the Pew Research Center found that “experts are far more positive and enthusiastic about AI than the public” and “far more likely than Americans overall to believe AI will have a very or somewhat positive impact on the United States over the next 20 years” (56 percent vs. 17 percent). And perhaps most glaringly, 76 percent of experts believe these technologies will benefit them personally rather than harm them (15 percent).
The public does not share this confidence. Only about 11 percent of the public says that “they are more excited than concerned about the increased use of AI in daily life.” They’re much more likely (51 percent) to say they’re more concerned than excited, whereas only 15 percent of experts shared that pessimism. Unlike the majority of experts, just 24 percent of the public thinks AI will be good for them, whereas nearly half the public anticipates they will be personally harmed by AI.
And as someone who has extensively set up such systems on their home server… yeah it’s a great google home replacement, nothing more. It’s beyond useless on Powerautomate which I use (unwillingly) at my job. Copilot can’t even parse and match items from two lists. Despite my company trying its damn best to encourage “our own” (chatgpt enterprise) AI, nobody i have talked with has found a use.
AI search is occasionally faster and easier than slogging through the source material that the AI was trained on. The source material for programming is pretty weak itself, so there’s an issue.
I think AI has a lot of untapped potential, and it’s going to be a VERY long time before people who don’t know how to ask it for what they want will be able to communicate what they want to an AI.
A lot of programming today gets value from the programmers guessing (correctly) what their employers really want, while ignoring the asks that are impractical / counterproductive.
You’re using it wrong then. These tools are so incredibly useful in software development and scientific work. Chatgpt has saved me countless hours. I’m using it every day. And every colleague I talk to agrees 100%.
I’ve found it primarily useless to harmful in my software development, making the work debugging poorly-structured code the major place that time is spent. What sort of software and language do you use it for?
Then you must know something the rest of us don’t. I’ve found it marginally useful, but it leads me down useless rabbit holes more than it helps.
I’m about 50/50 between helpful results and “nope, that’s not it, either” out of the various AI tools I have used.
I think it very much depends on what you’re trying to do with it. As a student, or fresh-grad employee in a typical field, it’s probably much more helpful because you are working well trod ground.
As a PhD or other leading edge researcher, possibly in a field without a lot of publications, you’re screwed as far as the really inventive stuff goes, but… if you’ve read “Surely you’re joking, Mr. Feynman!” there’s a bit in there where the Manhattan project researchers (definitely breaking new ground at the time) needed basic stuff, like gears, for what they were doing. The gear catalogs of the day told them a lot about what they needed to know - per the text: if you’re making something that needs gears, pick your gears from the catalog but just avoid the largest and smallest of each family/table - they are there because the next size up or down is getting into some kind of problems engineering wise, so just stay away from the edges and you should have much more reliable results. That’s an engineer’s shortcut for how to use thousands, maybe millions, of man-years of prior gear research, development and engineering and get the desired results just by referencing a catalog.
I’ll admit my local model has given me some insight, but in researching more of something, I find the source it likely spat it out from. Now that’s helpful, but I feel as though my normal search experience wasn’t so polluted with AI written regurgitation of the next result down, I would’ve found the nice primary source. One example was a code block that computes the inertial moment of each rotational axis of a body. You can try searching for sources and compare what it puts out.
If you have more insight into what tools, especially more i can run local that would improve my impression, i would love to hear. However my opinion remains AI has been a net negative on the internet as a whole (spam, bots, scams, etc) thus far, and certainly has not and probably will not live up to the hype that has been forecast by their CEOs.
Also if you can get access to powerautomate or at least generally know how it works, Copilot can only add nodes seemingly in a general order you specify, but does not connect the dataflow between the nodes (the hardest part) whatsoever. Sometimes it will parse the dataflow connections and return what you were searching for (ie a specific formula used in a large dataflow), but not much of which seems necessary for AI to be doing.
I think a lot depends on where “on the curve” you are working, too. If you’re out past the bleeding edge doing new stuff, ChatGPT is (obviously) going to be pretty useless. But, if you just want a particular method or tool that has been done (and published) many times before, yeah, it can help you find that pretty quickly.
I remember doing my Masters’ thesis in 1989, it took me months of research and journals delivered via inter-library loan before I found mention of other projects doing essentially what I was doing. With today’s research landscape that multi-month delay should be compressed to a couple of hours, frequently less.
If you haven’t read Melancholy Elephants, it’s a great reference point for what we’re getting into with modern access to everything:
https://www.spiderrobinson.com/melancholyelephants.html
If you were too lazy to read three Google search results before, yes… AI is amazing in that it shows you something you ask for without making you dig as deep as you used to have to.
I rarely get a result from ChatGPT that I couldn’t have skimmed for myself in about twice to five times the time.
I frequently get results from ChatGPT that are just as useless as what I find reading through my first three Google results.