https://nitter.net/petergyang/status/1607443647859154946
Dumb robot who can't do math 🤝 Me a dumb human who can't do math
The singularity has been achieved.
Once again reminding everyone that ChatGPT is essentially a fancy, electronic parrot
Reminding everyone that when someone says AI, they usually mean a fancy, electronic parrot.
That doesn't stop some folks, including on Hexbear, pulling the reductionist take of "and so are we, especially you if you disagree." Something something meat computers something something meatspace get schwifty :galaxy-brain:
Even yesterday I was called an "NPC" here because I didn't like one not-an-alt wrecker's take. Such reductionist cognitohazards are everywhere and they generally benefit the ruling class by demoralizing and dividing the rest of us. :capitalist-laugh:
This post is a great example of a place where AI cannot match humans, and won't be able to for a very long time or maybe ever. Computers completely shit the bed on interpreting ambiguous text from context clues (more on that here. If the problem were submitted in proper formula format to something like Wolfram alpha it would solve it no problem, but asking it conversationally gives results like this.
Even if/when computers start successfully and reliably understanding ambiguous text from context clues, I contend that what I said would still stand. There will be reductionists that want so very badly to declare the chatbot or other treat dispenser "true" intelligence in a way that belittles human intelligence at the same time. Instead of accepting the additional hurdles that such a machine would need to do to get there (plausible, with sufficient time, I believe), it's easier for them to denigrate human intelligence to try to rhetorically pull it down to the chatbot's level, now, for whatever reason. :lea-why:
Finding or making people that are more credulous and gullible is, technically, a way to make a machine more easily pass that test. :think-about-it:
Even yesterday I was called an “NPC” here because I didn’t like one not-an-alt wrecker’s take.
I always assumed you were the DM.
Not always, but DMs do have to play the part of every NPC the party encounter. :edgeworth-shrug:
Something something meat computers
That take is always extra frustrating because human brains are not computers. In fact we still have absolutely no idea what human consciousness fundamentally is.
That doesn’t stop some folks, including on Hexbear, pulling the reductionist take of “and so are we, especially you if you disagree.” Something something meat computers something something meatspace get schwifty
There was one meatspace bro on here who said that because of "meditation" that they agreed that we were basically giant meatballs or something like that. Honestly sounded like it came straight out of the Sam Harris subr*ddit...
Someone pointed out that this stuff is probably better viewed as a replacement for search engines than for people. It mostly outputs garbage, but also sifts through the garbage results google finds to give you the content of whichever stackoverflow page is relevant probably quicker than you can. And once that's aggregated a competent human can make sense of it.
so we add a bunch of natural language processing to search on the query side and now it's harder to look up specific things so the boys down in engineering cooked up this to pluck out the results i could've gotten 15 years ago with some knowloedge of booleans and regular expressions?
progress is so cool
This lets you customize results better. For example let's say you ask it to provide a snippet of code to do X. You can then provide it instructions to tweak it slightly
Earlier you would search online and find code to do X but you would need to know how to tweak it
i would simply get both the snippet and the tweak from stackoverflow
Eh, it's a bit better than a parrot. It gives you, what is essentially, the most statistically likely prediction (give or take) of what all the data it's fed suggests the answer should be. It's more like a really, really good probability machine capable of working with an insane amount of data. The most interesting thing about the chatgpt, I've found, is the size of its "short term memory" and the ability to prime it for the responses you're looking for. Last night I asked it if it was familiar with visual novels, and after it spit out some basic summary, I had it write a 3 act visual novel outline for the story of John Brown's raid on Harper Ferry. Then I asked if it was familiar with JSON structures and how they 're used by visual novel engines, and got it spit out the story outline, with a bit of extra interactions, as a JSON; Just not one that works with the engine I'm trying to use. I'm curious to see if I can teach it the json structure of a specific visual novel engine.
You can call it a parrot all you want, but you're missing out on the opportunity that this could be a big ol' Leftist agitprop slop machine.
this could be a big ol’ Leftist agitprop slop machine.
Who owns that machine right now? :cap-think:
Yes it does, actually. Just like each and every time in the past where some beneficent and magnanimous :porky-happy: seemed superficially generous up front, the time will likely come where that generosity dries up because of economic necessity and the "good" output becomes worse and worse over time.
There are a bewildering number of examples of this, but I'll use one of the biggest ones in a similar tech field from recent history: Google. It used to actually be a good, even great, search engine. And it was free. And people said the machine's owner did not matter because the output was good. :capitalist-laugh:
Of course we’re always going to have to adapt as capital catches up, that doesn’t mean these contraptions can’t have value in the short-term.
I have no disagreement over that particular point, except to say "adapt" is not the same thing as "submissive acceptance of what capital can do and will inevitably try to do with it."
:reddit-logo: was never good, especially because it swallowed up the userbase of countless smaller competing forums and was a near-monopoly of its kind. Even the "value in the short-term" has already been squeezed with cryptocurrency grifting attempts, deliberately worse interface and layout decisions designed to drive more "engagement," and clunkier and jankier everything except the ever more streamlined and more efficient monetization and surveillance/data-collecting systems that the ruling class benefits from.
Sadly, your jobs are not safe because this chatbot is who they're putting in charge of grading papers.
still laughing to myself over users here posting snippets of chatgpt-generated 'theory'
:so-true: "no, you see, I told the robot to use a Marxist framework!!"
i tried to ask it how catgirls fit into marxism. it claimed they don't. that's how i know i can't trust it for theory.
That one scene in The Thing where kurt russell pours his liquor into a computer after it beats him at chess
Hey so it's fun to shit on this thing but remember that it is still extremely powerful, a couple examples include having it write compellingly human product reviews for SEO and complete entire middle-school level essays, meaning overworked teachers will be grading AI generated work while students learn nothing
