Comment on I'll never forget how badly the term "AI" nosedived into being uncool
You are viewing one comment and its replies. View all comments · View the comment above
You are viewing one comment and its replies. View all comments · View the comment above
Comments
It definitionally is AI. It’s not the SciFi version or conscious or aware, etc. But it’s AI, and you need to provide a basis or evidence to say it isn’t.
I’ll go first: LLMs are built using artificial neurons that were built after modelling biological neurons, and they are trained by feeding large amounts data into the model to tune the weights. Similar to how biological neurons learn and tune the neural connections (weights) by learning from input.
Artificial neural networks share similarities (not all, not most) with biological neural networks.
I don’t really disagree, I just really like this comic:
https://feddit.org/pictrs/image/2de6642e-6c3c-4e08-95a7-f6586a5f9626.png
https://xkcd.com/1838/
They function in a way that is likely somewhat analogous to our language processing, but without an underlying ability to reason I think it would be a mistake to consider that “intelligence”.
It is surprising just how convincing the models can be sometimes given that limitation, but the unreliable behavior and jarring disconnects really put that lack on full display.
I’d argue a better term for them is “chatbot”.
If what you said is true, then half of the population are also in that category.
Have you heard what some people say and do? I’m not entirely sure there is a single intelligent thought going on in there, just an organic chat bot.
Lol, I have wondered if this might be a reason so many management types seem to fall for the propaganda. Like, “Autocomplete trained to convince an average person! It thinks like I do!”.
That said, people at least have the capabilities of reason and empathy, even if some choose not to use it.
AI really does “think like management”. Both are this weird least common denominator, what is popular, go with the flow, don’t think, just do it and move on type of system. So yeah, you end up with “good enough, ship it!” Mentality on everything.
I think you’re overestimating the human ability to “reason”. I would argue that the overwhelming majority of human “reasoning” is more accurately described as “sophisticated mimicry”, with little actual “intelligence” behind it.
“unreliable behavior” and “jarring disconnects” are far from absent in society.
What do you think human reasoning is mimicking?
Sure, many people don’t do it well, but they have the capacity to do so. We are constantly modeling scenarios in a much more complex and detailed way than next word frequency prediction. I wouldn’t be able to function without an automatic inner monologue otherwise.
Prior work. The same thing that LLMs are mimicking. Memes, multiplication tables, scientific papers, political soundbites, jokes, sonnets, movies, comic books…
Which has the higher capacity for reasoning: the dumbest person you’ve ever met, or the most capable LLM you’ve ever heard of?
The person. Try it yourself!
What’s funny to me is this whole argument over the term “AI” being used and whether this artificial thing we created is “intelligence” or not and I still honestly don’t know if most humans are intelligent or not.
Half of humans would probably fail the Turing Test.
The rebut I would give is the randomness of biological systems that is built into it. I know there are analogues for it but it’s clearly not yet keeping up with that.
That’s more to do with the hardware itself. Nueral tissues seems to function with a “warm” pseudo-quantum processing (if certain unverified theories are true).
We’re also emulating analogue calculations (neural nets) on digital hardware. I’m really excited to see what these analogue accelerators can do once they take off. Their supposed to be faster than GPUs for LLM inference while using 1/10th the energy.
Sign in to comment. Sign in