• 1 Post
  • 425 Comments
Joined 2 years ago
cake
Cake day: September 24th, 2023

help-circle

  • Private or obscure ones I guess.

    Real-world (macro) benchmarks are at least harder to game, e.g. how long does it take to launch chrome and open Gmail? That’s actually a useful task so if you speed it up, great!

    Also these benchmarks are particularly easy to game because it’s the actual benchmark itself that gets gamed (i.e. the code for each language); not the thing you are trying to measure with the benchmark (the compilers). Usually the benchmark is fixed and it’s the targets that contort themselves to it, which is at least a little harder.

    For example some of the benchmarks for language X literally just call into C libraries to do the work.



  • It has no memory, for one.

    It has very short term memory in the form of it’s token context. Especially with something like Meta’s Coconut.

    What makes you think that it does know its in a conversation?

    I don’t really. Yet. But I also don’t think that it is fundamentally impossible for LLMs to think, like you seem to. I also don’t think the definition of the word “think” is so narrow that it requires that level of self-awareness. Do you think a mouse is really aware it is a mouse? What about a spider?




  • And how do you know LLMs can’t tell that they are involved in a conversation?

    Unless you think there is something non-computational in the human brain, then you must accept that computers are - in theory - capable of thinking. With the right software and sufficiently powerful hardware.

    Given that truth (which I think you can only avoid through religion or quantum quackery), you can’t just say “it’s only maths; it can’t be thinking” because we know that maths can think.

    Do LLMs “think”? The definition of “think” is wooly enough and we understand them little enough that it’s quite an assertion to say that they definitely don’t.






  • LLMs can’t think - only generate statistically plausible patterns

    Ah still rolling out the old “stochastic parrot” nonsense I see.

    Anyway on to the actual article… I was hoping it wouldn’t make these basic mistakes:

    [Typescript] looks more like an “enterprise” programming language for large institutions, but we honestly don’t have any evidence that it’s genuinely more suitable for those circumstances than the regular JavaScript.

    Yes we do. Frankly if you’ve used it it’s so obviously better than regular JavaScript you probably don’t need more evidence (it’s like looking for “evidence” that film stars are more attractive than average people). But anyway we do have great papers like this one.

    Anyway that’s slightly beside the point. I think the article is right that smart people are not invulnerable to manipulation or falling for “obviously” stupid ideas. I know plenty of very smart religious people for example.

    However I think using this to dismiss LLMs is dumb, in the same way that his dismissal of Typescript is. LLMs aren’t homeopathy or religion.

    I have used LLMs to get some work done and… guess what, it did the work! Do I trust it to do everything? Obviously not. But sometimes I don’t need perfect code. For example recently I asked it to create an example SystemVerilog file for me utilising as many syntax features as possible (testing an auto-formatter). It did a pretty good job. Saved some time. What psychological hazard have I fallen for exactly?

    Overall, B-. Interesting ideas but flawed logic.





  • I don’t think this is a very interesting article. We already know AI suggests nonsense a lot of the time. That in no way demonstrates that it is net-negative. In my experience it’s a net positive even accounting for the times it gets things wrong.

    Yes you do have to review its code closely. News at 10.

    It is kind of funny that they picked an example where it made an obvious mistake for their hero image though.




  • I agree. I think it’s driven by fear. I get it. I’m slightly afraid I won’t have a job in 10 years (or at least a much worse paying one)…

    I’m still a much better programmer than AI today. But I don’t cope with the fear by deluding myself into thinking that AI is useless and will stay useless.

    The feels a lot like portrait painters saying that photography will never amount to anything because it’s blurry and black and white.