Comment by limflick
12 hours ago
Auto-complete on steroids, is still my favorite analogy for AI. I don't mean that in a negative way either. Autocomplete is very good, but that never stopped me from learning English grammar and spelling.
12 hours ago
Auto-complete on steroids, is still my favorite analogy for AI. I don't mean that in a negative way either. Autocomplete is very good, but that never stopped me from learning English grammar and spelling.
Quite right. I'm worried about the impact that LLMs will have on the learning process, especially in programming, but also in writing. Programming and writing are both skills that seem simple, but take an absolutely staggering amount of practice to master.
Think about how much your own writing (and programming, if you were lucky enough to start early) evolved from, say, age 12 (when a lot of smart kids start to tackle 'real' books) to age 18 (when you supposedly have a good enough education for 50% of work in most countries) to age 25.
All of that evolution is a direct result of one thing: practice! But with a magic answer box available in everyone's pocket, it'll take truly Herculean effort from a learner to actually grind through the practice instead of just cheating for an answer. I really worry how much an LLM user will actually comprehend their own code or even prose; if you've scarcely written a line of code, how can you really understand what's going on in a debugger? If you haven't done the legwork of writing essays and constructing coherent arguments and comprehending grammar, how will you ever communicate effectively?
Maybe I'm just a dinosaur and these kids will sail a whole level of abstraction above my own understanding of writing and programming, much like how my own generation preferred Python to C, and how the previous generation evolved from assembly to C/BASIC/etc. But then I come back to those missing fundamentals, that empty mental model. It's not like my English or CS teachers had me grind through essays and implementing linked lists and Djikstra's Algorithm for pure busywork. They did it because practice is the only way to truly immerse a student in a practical subject. Maybe it'll work for programming, as long as LLMs get good enough that you can always ask them to fix low-level errors for you? But it seems unlikely to work in prose. And even those generational programming jumps I mentioned (assembly to C to Python) were lossy; most kids I went to school with would be absolutely useless writing C code, and even as a bit of a dinosaur I'm pretty awful at even debugging assembly.
Like you said: you still need to learn grammar and spelling. And I suspect a whole skill tree of other fundamentals!
One angle I'm exploring, as a non-dev who nonetheless works in tech, is using Claude as a professor. Make learning timelines for me for Leetcode, break it down in phases, start with theory, ask me questions, then give me a coding challenge. Save that to an html artifact I can export and read on my phone.
It still gets things wrong, I can tell as I get through problems.
But it was either that or that dreary 'Cracking the Coding Interview' book. At least I'm learning fundamentals by asking question after question and making it track the concepts I had trouble with.
That's one use. Will most people use it to learn? Probably not. But most people are ... most people.
Yup, I used to believe that people would all use the Internet to educate themselves, and we all know how that turned out (loads of people did, but the majority didn't).
The way I think of it has evolved a lot over the last 5 years. At this point I think human brains probably do something analogous to next token prediction when we think. For all the hype, I think LLMs are actually more, not less, intelligent than that average person realizes. I think it’s legit, actual intelligence, not just “artificial” intelligence. That may be a hot take but it’s just my perception.
> At this point I think human brains probably do something analogous to next token prediction when we think
That's reasonable, but it doesn't mean that LLMs are close to being brains.
For a start, when humans think/talk, we often think ABOUT something - whatever is swirling about in our mind, or what we are currently seeing/feeling/etc. An LLM generating tokens/words is doing so only based on it's weights and the word sequence it is currently generating ... the human parallel would be more like a rapper spitting out words based on prior words, essentially on auto-pilot, or when we get triggered into spitting out stock phrases like "have a nice day".
If you want to compare an LLM to a human brain, it's basically equivalent to our language cortex if you ripped out all the external connections and ripped out all the feedback paths that make it capable of learning.
Of course there is a lot more to our brain than just our language cortex, but that alone should make you realize there is no real comparison beyond the fact that our language generation is also going to be based on prediction, and partly auto-regressive.
We have spatial / quantitative and social / emotional aspects in our intelligence that are not at all like next token prediction.
If LLMs had shame, they'd surely not repeat mistakes (in the same context window) as much as they do.
Having shame would require the LLMs to actually be able to recognize mistakes they make.
People love to put a lot of meaning on what an LLM responds with when asked why it made a mistake, but it's critical to remember that the answer to that prompt is just another series of probabilistic tokens, and has no actual relation to how the error happened.
2 replies →
It's language. Language itself is the thing that makes us smart in the unique way that we are among the other animals, and it weirdly turns out to be transferable to machines to at least some degree.
At least 50% of humans have no "inner voice" and are not thinking in the same way as you. Many animals like dolphins, dogs, rats, crows are also very intelligent yet appear to only have primitive language capabilities.
A lot of human intelligence is really societal rather than individual, based on knowledge transmitted down through generations by writing (the real enabler). If you take that away then what you are left with is something more like an isolated hunter-gather tribe.
4 replies →
I think you’re right but I wanted to note that we are discovering that other animals also have language.
1 reply →
> I think it’s legit, actual intelligence, not just “artificial” intelligence. That may be a hot take but it’s just my perception.
You might be redefining words here; there isn't a form of intelligence that isn't actual intelligence. It is all actual intelligence. Artificial in this context means it is something we're creating in a lab. LLMs can't avoid being artificial intelligence. The meaning of "AI" is to artificially create actual intelligence.
average person is absolutely awful judge on anything you put in front of average person tho.
And if anything, average AI user is vastly overstating how good/useful it is. Papers about it pretty much always show huge gap between "productivity person thinks they are achieving" and "actual growth of productivity"