Ha, I’ve done the same thing. GPT-4 is like a super nice intern that knows a ton but doesn’t get the big picture.
617 karma · joined June 4, 2012
My rule for students is basically, "If you have the option to be re-specify the thing you're talking about, do it." That's a solid rule for standardized tests, and usually applies to everyday writing (I'll caveat that this rule clashes with "the simplest answer is the right answer", so it is dependent on the actual sentence rather than an "all-or-nothing" rule).
Other common mistakes are those you hear about in middle-school (that's not a knock on anyone; rather, I say that to prove how long-ago it was that most of us ever reviewed common grammar and writing rules):
"Let's eat, Grandma!" vs. "Let's eat Grandma!"
Tying this back to GPT, I've read (and seen) folks write without any punctuation whatsoever. I can't speak to how GPT truly handles that, but if it's anything like "normal writing" and understanding, then punctuation is hugely important.
I have a side business for ACT/SAT prep. I teach English, and often have to remind students about sentence structure and word flow. For example, I can't say "My mom, my grandma, and I went to her favorite store" -- in that example, there is no clear way to know who "her" is.
Similarly, I see many people claim GPT-n is "dumb", yet when I see their prompt, I realize it was a bad prompt. There are clear logical inconsistencies, there is inadequate information, and there is confusing word usage.
I've been astounded by GPT-4 and have nearly 5x-ed my productivity with it (for coding and for parsing documents). But I think my experience is a result of my habitual "standardized testing" writing style, while others' poor experience is a result of their more layman writing style.
"I'm most lonely when I'm alone with my wife" -- that is not healthy or good. I don't think your situation is applicable to the OP.
As a counter, I feel most comfortable when I am alone with my wife. She is my best friend.
Should I forgo my modern living entirely because of my carbon footprint? Where do you draw the line?
People continue to claim LLMs are just parrots. But then there is never a solid rebuttal to how humans are different.
I'm not saying we don't, but I am saying these sorts of arguments always leave so many begging questions. "What does it mean to think abstractly? What does it mean to think? What does it mean to reason?"
“It’s simple ! Just tokenize the context and allow differentiated hash maps to best map the embeddings matrix. Duh!”
And I just want to ask if any of these people every met any real child.
Imagine you are 14. You hear the climate isn't doing too hot, but then you pull out your phone and see your classmate getting tons of attention for some post he made on social media. Meanwhile, your own post has barely any likes.
Which is going to bother you more? Climate change, or the social media sadness?
But also, I'm not so sure scientists "don't have the time" but rather, "literally cannot" know with regard to an LLM's "insides".
Also, great bit here:
"some tech communities are naturally much more _about_ bringing forward the best parts of the technological spirit of the past, and less about celebrating what's just arrived."
I try to be wary of this as well. I'm often reminded of the famous Dropbox HN post, where one of the top comments said "why Dropbox? Just make it yourself".
Not sure how relevant that is, but some AI discussion does indeed remind me of that.
I will agree that it's by no means Star Trek level, but it is about the most fascinating piece of technology I've seen in my lifetime.
And yes, that sounds silly! But therein lies my reason for writing this post.
Are gamechangers known to be such upon arrival? (That's a real question -- I might argue they aren't known immediately. )
But that's beside the point:
How do people not extend the timeline a bit? It often feels like the release of the internet, but instead of people going, "Holy crap, this is going to be huge!" we have a majority saying, "meh, it only connects like 12 different universities."
As though progress somehow deadstops right where we are.
The "inside" of an LLM is a black box in that we don't know how an input arrives to an output. We know some general ideas ("it predicts the next thing"), but not the detailed route to get to the output.
Similarly, we do not know how human minds work. How does a thought arise? How does a new idea arise?
I would not argue that an LLM is identical to a human brain. But I do think there's some interesting discussion to be had on that topic.
It's as though we have two black boxes, and so many folks are willing to say, "One of these is definitely, certainly not like the other black box!"
And I don't see where that confidence comes from.
That comparison, to my silly mind, is astonishing to think about. And I've yet to see why I should (or could) dismiss the notion of similarity.
Or, simply using GPT-4; that's quite astonishing to me. I'm not a stellar coder by any means, but I've quite literally 5x-ed my productivity with GPT-4.
"Well, you must not be working on novel problems." You're right! But is "solving novel problems" a requirement for discussion now?
I will try another approach:
I typically would label the HN crowd as above-average when it comes to understanding technical progress. I would not call myself that; I'm just a software dev that enjoys tech and happens to also enjoy learning about the human brain.
And my question stems from the fact that I see all these smart folks initiating conversations, like "it doesn't understand anything", and then whenever a follow-up is presented ("what does it mean to understand?"), it's silent.
That's a common theme I've seen these last 3-4 months. And I truly feel like I'm missing something.
Even assuming that definition, it begs the question of, "what is creativity?"
I struggle to see how GPT-4 is not intelligent by any definition that applies to a human.
The constant downplaying and denial is unreal.
“Why doesn’t a person like him have better data? This isn’t cool for AI; it’s simply a bullshit problem that should’ve never existed.”
“How does he know the data was accurate? These things are just hallucinating parrots.”
Isn't a normal human just as likely to be inaccurate or make mistakes? The world isn't filled with perfect data in every format. Good grief.
Perhaps I get worked up and comment here in hopes that the AI overlords will view me favorably based on my online comments. My real frustration, though, is with the significant number of "usually smart" people who continue to downplay such incredibly cool technology.
Do I know it’s “true” that the acceleration of gravity is 9.8 meters per second squared? I’ve never experienced the actual tests. I’ve never vetted that out. I read about it once. And I “parrot” that fact all over the place! And it’s helped me solve some silly physics problems I’ve explored for fun. It’s helped me do some fun predictions with simulators.
That’s probably not the best example, but it proves the point I’m trying to make:
Why is there this focus on “LLMs can’t understand”, when A) I’m not sure we can even define “understanding” and B) why does it matter ?
And I’ve said this across so many threads now but—
Can anyone explain to my why we should be so confident that humans aren’t just parrots of sorts themselves? No, I’m not saying we are similar. But I push back on the confidence with which people so solidly proclaim “the LLM’s blackbox is nothing like a human’s blackbox!”
Everyone shrugs and says, “nope, humans are different”. I’ve commented about 100 times recently asking for detail as to how human language / thought works, yet have not seen an answer.
We don’t know that humans themselves aren’t just prediction machines. Yet , humans are insanely capable! And thus the same might apply to an LLM.
It’s hard to strike a balance between having excited, rational, discussion and not coming across like a religious AI nut.
The thing I find interesting in the current LLM conversation is how much it’s opened up conversations around “what is knowing? What is consciousness?”
Does it matter, not philosophically but utility-wise, if something can give the illusion of intelligence or “knowing”?
And to go on something you said about taste— this is the age-old conundrum that still applies to humans, not just LLMs. I will never know if the color green to me is the same color as it is to you; our subjective experience may be entirely different, yet we both use the same language to describe what we experience.
And that language does the job just fine, even without the experience itself factored in.
I’m not claiming LLMs are sentient. I’m not claiming they are even similar.
What I am pushing back against is the confidence with which people so blatantly claim we are dissimilar.
It’s an important distinction, and I’ve yet to see solid evidence to suggest it’s a point we shouldn’t even explore.
What I see so often is comments stating things like, “an LLM is just pattern matching” or “it’s a prediction machine”.
And I’m not arguing that’s not true; what I’m arguing is how can anyone say a human is inherently different ?
I admit that I’m taken with these latest advancements. It’s why yes, you see my tech-bro crazy-person comments in these threads a lot! But I’m genuinely fascinated by this stuff.
I have had a huge interest in the human mind for a decade. I’ve read the works of folks like Anil Seth and others who work in cognitive and computational science, and I’m increasingly intrigued by the things we might learn about our own selves via these technological advancements.
Again, I’m not claiming sentience for an LLM, or anything that says “we are the same as an LLM.” I’m simply trying to argue that our own minds are a black box, as are how an LLM arrives at an output; it’s the confidence of folks to claim they know what’s inside either that I’m pushing back against.
But doesn’t it imply that once we have something that can “see” its environment, hear nearby sounds, and have some sort of haptic sense — wouldn’t that be AI, then? Given your definition?
If, of course, such a thing used all that input to “create” concepts like its physical place within an environment, what may or may not come next given the interplay of those inputs, and how to navigate the environment based on all that.
But then I’d say, “GPT-4 can handle image data. And what might come in the future with more modalities?”