korrents

korrents · video

Not this again

NeetCode · 21m · youtube.com ↗

20 korrents from this recording

NeetCode did not write this page.

Every claim below is a statement made in this recording, quoted word for word and linked to the second it was said, so you can hear it rather than take our word for it. The wording comes from the transcript published alongside the recording; the sentence above each quote is our reading of the claim, not their wording.

  1. 2:03 · watch on youtube.com↗

    I've noticed a trend, and we kind of all have, where words just really don't mean anything anymore, and I thought this was supposed to be programming, where things have like a precise meaning, and words matter, definitions matter.
  2. 1 min later
  3. 2:39 · watch on youtube.com↗

    how can we expect that coding is solved implying that you can just tell an agent what to do and it'll completely understand everything and then just do everything and it'll do it precisely when we can't even agree on what words are supposed to mean anyway.
  4. 2 min later
  5. 4:41 · watch on youtube.com↗

    nobody can predict the future. I know that I say that like every single time, but it matters because unless Boris is talking about something that's already happened, something they're already experiencing inside of anthropic, maybe with models that we don't have access to, then you have to just be skeptical of anybody trying to predict the future.
  6. 2 min later
  7. 7:02 · watch on youtube.com↗

    And for me, it felt more like scavenger hunt programming. It wasn't really about putting ideas together. It was just about finding the information that you needed to know. And I'm really glad that models can do that part now.
  8. 1 min later
  9. 7:42 · watch on youtube.com↗

    Fable isn't really going to ask you like a follow-up question like a normal engineer would. It's just going to implement whatever it thinks is closest to what you're saying, even if you're not precise.
  10. 8:02 · watch on youtube.com↗

    I think most people who are doing agentic programming, including Boris, are probably somewhere in between that. you're not just giving super vague things, but you're also not going line by line telling it, you know, what to do.
  11. 1 min later
  12. 8:49 · watch on youtube.com↗

    It's not like whatever an LLM can do, what humans can do, it's not really a onetoone comparison. It's they're kind of complimentary.
  13. 9:07 · watch on youtube.com↗

    I think most people would agree that LLMs are not going to be enough to completely mimic the human human brain. I don't think we're going to get there and that's like a simple reasoning for it. Maybe I'm wrong.
  14. 9:20 · watch on youtube.com↗

    But whatever the LLM can't do without people, that's what people are probably going to end up doing.
  15. 9:26 · watch on youtube.com↗

    And even now there's still tens of millions of programmers employed. If companies really didn't need them, they would not be employed.
  16. 2 min later
  17. 11:06 · watch on youtube.com↗

    I think coding isn't something that really can be done just in isolation. Like there's so much more context that goes around that even to know just like what to code.
  18. 2 min later
  19. 12:43 · watch on youtube.com↗

    That's why I think it's kind of absurd that we're going to get to a point where code doesn't even need to be reviewed anymore, even by other LLMs.
  20. 1 min later
  21. 13:18 · watch on youtube.com↗

    Agents are perfectly capable of doing that, but they don't tell you to do it. You have to know to do that.
  22. 2 min later
  23. 15:31 · watch on youtube.com↗

    I feel like that would create more noise than like value cuz I mean you can end up with like a hundred of these but somebody still has to go and review them and test them and see if they're actually valuable.
  24. 2 min later
  25. 17:50 · watch on youtube.com↗

    I think there's no risk in having like just taking all the context that you know of like what happened with the alert giving that information to an agent because agents can more quickly than humans read things right.
  26. 1 min later
  27. 18:24 · watch on youtube.com↗

    when I was working at Google, most alerts that we were getting were just noise. They were non-actionable. Nothing we could really do about it. And I think that's a perfect instance of having like an agent do something.
  28. 1 min later
  29. 19:44 · watch on youtube.com↗

    And so, like, again, a non-coder just wouldn't even look at that. Like, if you don't look at these things, you miss a lot. Like, and the more you look, the more issues that you see.
  30. 1 min later
  31. 21:11 · watch on youtube.com↗

    it's not thinking the way you think it is. It's like it's a text engine. It's like it's autocomplete.
  32. 1 min later
  33. 21:42 · watch on youtube.com↗

    But the LLMs just can't for whatever reason sometimes suggest the obvious solution. They want to give you more complicated ones.
  34. 21:49 · watch on youtube.com↗

    So I don't know. I just I feel like the details matter. I feel like understanding things, thinking about things does matter.