Yknow, at the time I figured this guy, with his background and experience, would be able to distinguish normal from abnormal LLM behavior.
But with the way many people treat GPT3.5/GPT4, I think I’ve changed my mind. People can know exactly what it is (i.e. a computer program) and still be fooled by its responses.
I get what you mean, but I believe it’s more productive not lumping a neural network (inference model), with much of the “logic” coming from automated/self-training, into being “just a computer program”. There’s historical context & understanding of a “program” where a human actually designs & knows what IF-THEN-ELSE type of logic is executed… understanding it will do what it is ‘programmed’ to do. NN inference is modeled after (& named after) the human brain (weighted neurons) and there is both a lack of understanding all (most!) of the logic (‘program’) that is executing under-the-hood, as they say.
Note: I’m not at all saying that GPT 3.5/4 are sentient, but rather that it’s missing a lot of the nuance, as well as complexity, of LLMs by referring to them as simply being “just a computer program”.
If you ever wonder if the machine is sentient, ask it to write code for something somewhat obscure.
I’m trying to run a Docker container in NixOS. NixOS is a Linux distro known for being super resilient (I break stuff a lot because I don’t know what in doing), and while it’s not some no-name distro, it’s also not that popular. GPT 4 Turbo has given me wrong answer after wrong answer and it’s infuriating. Bard too.
If this thing was sentient, it’d be a lot better at this stuff. Or at least be able to say, “I don’t know, but I can help you figure it out”.
I think this is a huge problem with current AIs is that they are forced to generate an output, particularly in a very strict time constraint. “I don’t know” should be a valid answer.
I’m more talking about hallucinations. There’s a difference between “I’m not sure”, “I think it’s this but I’m confidently wrong”, and “I’m making up bullshit answers left and right”.
Yknow, at the time I figured this guy, with his background and experience, would be able to distinguish normal from abnormal LLM behavior.
But with the way many people treat GPT3.5/GPT4, I think I’ve changed my mind. People can know exactly what it is (i.e. a computer program) and still be fooled by its responses.
I get what you mean, but I believe it’s more productive not lumping a neural network (inference model), with much of the “logic” coming from automated/self-training, into being “just a computer program”. There’s historical context & understanding of a “program” where a human actually designs & knows what IF-THEN-ELSE type of logic is executed… understanding it will do what it is ‘programmed’ to do. NN inference is modeled after (& named after) the human brain (weighted neurons) and there is both a lack of understanding all (most!) of the logic (‘program’) that is executing under-the-hood, as they say.
Note: I’m not at all saying that GPT 3.5/4 are sentient, but rather that it’s missing a lot of the nuance, as well as complexity, of LLMs by referring to them as simply being “just a computer program”.
it’s dismissive & rude for you to call it “fooled” that he came to a different conclusion than you about a subtle philosophical question
If you ever wonder if the machine is sentient, ask it to write code for something somewhat obscure.
I’m trying to run a Docker container in NixOS. NixOS is a Linux distro known for being super resilient (I break stuff a lot because I don’t know what in doing), and while it’s not some no-name distro, it’s also not that popular. GPT 4 Turbo has given me wrong answer after wrong answer and it’s infuriating. Bard too.
If this thing was sentient, it’d be a lot better at this stuff. Or at least be able to say, “I don’t know, but I can help you figure it out”.
I think this is a huge problem with current AIs is that they are forced to generate an output, particularly in a very strict time constraint. “I don’t know” should be a valid answer.
At this point I’m probably not sentient either
I’m more talking about hallucinations. There’s a difference between “I’m not sure”, “I think it’s this but I’m confidently wrong”, and “I’m making up bullshit answers left and right”.
Are we? Do we have free will or are our brains are just deterministic models with 100T parameters as mostly untrained synapses?