Today's manmade horror: Bazingas on twitter are amazed that glorified chatbots like GPT-3 can produce readable text that is nominally about whatever topic you request. Say they just spent 20 minutes learning about physics. Say that this makes all existing forms of education obsolete. Say that teachers are no longer needed.
Directly below this someone else posted a prolonged dialogue where the GPT-3 bot calmly explained why a peregrine falcon is the worlds fastest marine mammal. When corrected it said that the sailfish is the fastest marine mammal, explaining that sailfish are warm blooded and have fur.
Bazingas think they can replace teachers (and everything else) with their half-baked machine learning systems because they're so ignorant and uneducated about anything outside their direct field of expertise that they don't actually know what any other disciplines do, or why they do what they do, or even how to evaluate when they're being fed bad knowledge.
Link for dunking.
One of my jobs involves training a model for something very specific in industry. This is what AI is good at: it's just an accelerated statistical analysis framework that'll pick out a signal and give you pretty reliable results when given really specific parameters. Trying to turn that back around and make generalized frameworks that can be "intelligent" just isn't feasible, not for a long ass time. Even then, I'm skeptical that humanity will remain stable enough to produce the kind of chips that would be required.
I'm not saying true intelligence from artificial sources is conceptually impossible in the future, but the people that look at chatbots and say "this is it" are reductionist about what human intelligence is and what it can do and often have contempt for people in general.
Not impossible, but generalized intelligence that can just take a bunch of (often contradictory) information and be sensible to talk to is what I'm skeptical will be achieved before the next dark ages lol.
Pretty capable chatbots that are as accurate as a high school teacher about specific topics are essentially possible today, (like the "teach me Python" guy's example in the thread on the bird site which is probably fake anyway because that's easier). It just requires a ton of parameter definition and manual intervention, forcing users to use specific syntax that probably means a searchable ebook would've been a better use of the subject matter experts' time that you'd need to properly train it with feedback.
A lot of articles are like "Look at this amazing game changing thing I did with AI" and the amazing thing is replicating, say, a web page with a calorie calculator that provides meal plans, except that with the AI output you have to go manually check that the numbers are right because the AI is just cribbing random information that is statistically similar to the prompt.
There is so, so, so much contempt for the process and skills and creativity of artists going around. "Akshually humans just observe things and then mash those things together to make new things, exactly like this math problem does!" Like jesus christ how do you argue with something like that?
I don't know, but they tend to also say euphoric things like "humans are just meat computers" and tend to be insufferably smug. :very-intelligent:
Ugh. I am 100% a meat computer guy. I don't think there's a unitary self, I buy in to the idea that our "conscious" mind is mostly an illusion and the real thinking is done by hundreds of interacting non-conscious modules in the brain, I think (with evidence) that our perception of the world is extremely limited and inaccurate and propped up by so many different hacks and shortcuts that it's a wonder we can perceive anything, and I still think these guys are misanthropic shithead creeps.
Who don't understand what they're talking about.
Because whereas I read a very small number of neuroscience articles that I don't understand, they seem to read no neuroscience articles yet very confident in their ability to understand neuroscience.
I'm not making a special case for a soul or whatever straw argument hard determinists might set up for me. What I'm saying is that it's lazy and reductionist to take the human brain and compare it to binary coding. Neurons may have on-off switches but nonetheless have a lot of "maybe" for when they fire, so comparing human computation to binary coding is both reductionist and incorrect. Also, reducing the whole of human thought to chatbot comparisons is similarly lazy and reductionist. I extend that argument to what I've already seen among some online: that chatbots are basically human neural nets already (bullshit) and that people had no worth or right to exist over sufficiently elaborate chatbots made to replace them (fuck off, techbros).
A puppy may be a few cents of common chemicals and be nothing more than those chemicals in an ongoing biological process, but it's like comparing a puppy to a Tamagotchi and claiming that the Tamagotchi is in all meaningful ways the exact same thing as the puppy.
To your credit, you're not one of the insufferably smug eliminative and reductionist pop-nihilist types I'm talking about. I tend to get along with you even if I don't often agree with you.
I'd give it 400 years at least
deleted by creator
I've recently come around to the idea that building an AI from scratch is really dumb when we've got billions of perfectly functional brains to screw around with. Like sure, building an AGI from first principles would be cool, but we could also give me an extra arm and I think that would be much cooler.
If bazinga billionaires finance an intelligence being constructed, they get to decide what constitutes "friendly" for that intelligence. No wonder so many bazinga AI fantasies involve digital waifus that either obey them with the utmost devotion or "go evil" and kill them, implying there's nothing else for a true artificial intelligence to think about except senpai. :pathetic:
I meant in the dystopian sense