@simon One can object to many things about AI, but to think it’s *useless* is just willful ignorance.

@simon For me, 2024 showed that the power usage of inference *can* be reduced to ethically viable levels. But nobody cares, so still a net negative for humanity.

@simon Voted useless because I don't think consumers are willing to pay the real costs of these systems.

@simon I wanna see the results on bsky

@simon I fall along the lines of "it's about as useful as when I was attending workshops on it in 2018". I can see lots of practical applications for it, but nothing earth shattering, and I've yet to find a use for it I didn't foresee in 2018, though it is far easier now to implement.

@simon I think it is profoundly unsuited to many things it is being used for — and reading what you’ve shared over last year make it very clear that used carefully, understanding how it works, and with appropriate expectations it can be incredibly useful.

I’m used to being able to write incredibly complex programs to do very interesting stuff. When actively programming my mind is holding a huge amount of understanding. This is what I’m used to.

1/2

@simon

For me, it's not that it's not useful for *anything*, it's that it has been overhyped and overpromised to a ridiculous degree.

@simon not that it eas 100% useless, but that it was hypes for the wrong tasks.

Heck, I'm sure things like protein folding have greatly benefitted.

Code, though...

@simon Like any new tool, it takes a while to figure out how to use LLMs effectively. The unusual thing is the speed at which they are getting better. I’m writing code by describing what I want to ChatGPT and getting code that’s much better than I could write in a few seconds instead of giving up when I run out of time reading man pages or trying to figure out what library/function to use and what it’s called. Over the last year they’ve got good at a lot more languages.

@simon
I thought from the start that LLMs are potentially slightly useful on their own; and potentially quite useful as the language module of a larger system. And they'd need to become actual products in either case, which hasn't happened. That assessment hasn't changed.

But I also thought the hype around them is *vastly* overblown, driven by mistaken beliefs on what they actually do, in combination with an industry in desperate search of a new hit. That assessment hasn't changed either.

@simon after reading the comments I should change from “totally” to “slightly” useless. For example, a friend of mine uses it to write exam questions and thinking about it, bland and generic prose is actually a big plus there. But the costs (errors, security, author’s rights, environmentally) are out of proportion to an almost laughable degree.

Image and pattern recognition though. That’s going to save lives if it doesn’t already.

@simon
Went with useless because you didn't have a "harmful" choice :) (Yes, I do keep seeing cases where people somehow pry value out of it, I just think on average it still gets in the way.)

@simon I've never been in the "completely useless" camp, but the last year has taken me from "seems like mostly overhyped trash, but maybe I'm missing something" to "it's fundamentally unfit for most of the things it's being hyped for, and none of the people hyping it have a credible plan to fix that"

@simon when right wing billionaires market a mishmash of tech & sci-fi as a new capitalist religion & cure-all, which they have successfully forced into every product, even when clearly causing harm…

The abstract question "is this tech useful sometimes?" just isn't my top concern. I'm opposed to the way it's being deployed against people. I'm opposed to the vision of the future being sold as "AI".

@simon real answer is more detailed. It can be very useful for certain types of task, it can actively hinder some tasks, and people tend to defer to its confident tone when they should not. But it is also environmentally, financially and legally unsustainable in its present form and context.

@simon 1/2: I find AI deeply unscientific the way you interact with it. Knowledge in modernity took a turn epistemologically speaking from god to an agreement like relationship between theory and practice. Knowledge is not what agrees with someone or is true but what represents an agreement between theory and practice. All scientific work revolves around that. These sophisticated text generators are little more than lemma generators.

@simon 2/2: You can look up a mix of words that may be true believing to have gained insight without engaging into building your own model of something or performing your own experiment. Now if it were to force you to do just that... Its a shortcut to mediocrity. I usually use it for tedious tasks that I am unable to automate or ask it to scrutinize my premises, models or experiments but, to no surprise, I rarely find out anything that's not contained in a well curated article or dictionary.

@simon I don’t think AI’s useless. I think it’s being used for all sorts of things it’s not appropriate for because of hype, i think almost everyone who stands to make any money from that happening is annoying, and most importantly i think that it’s more damaging than any use is worth. Like, say, fossil fuels (which is not a bad analogy). This year has only multiplied the above sentiments more. But I suppose this means the poll doesn’t apply to me!