[–] 8 points 2 months ago

Several of those are fairly borderline. You would think Drew Devault of all people would understand that its ok to not be 100% squeaky clean.

I'm not saying that list doesn't include a lot of creeps, but maybe he shouldn't have included "donated to an organisation that includes someone who is antagonistic" and "is an average American Christian" along with the "fervent defender of paedophilia".

  • source
  • parent
  • context
  • [–] 1 point 2 months ago

    AI is not capable of novel or complex stuff,

    This seems to be both untrue and mostly irrelevant.

    The recent FLT proof is clearly complex and novel.

    But also a huge amount of programming is only barely complex and not really novel at all.

    The biggest issues I've seen with AI so far are:

    • The slop comments are just painful to read. I think this might mostly be Claude though, I recently switched to Codex and it's way less annoying.
    • It is so benchmaxxed to "succeed at all costs" it will often do stupid things just to make the test pass or whatever. To be fair a lot of humans are like that - "I dunno but it seems to work" attitude.
  • source
  • parent
  • context
  • [–] 1 point 2 months ago (1 child)

    Python is a poor choice IMO. It's just too slow, even for web servers. Everyone says "oh it's IO bound or database bound; Python doesn't matter") but in my experience it does.

    Better choices:

    • Go - very easy. Build in web stuff. Fast. Trivial to deploy.
    • Typescript, especially via something like Deno's Fresh. Much faster than Python, better type checking, and best of all you can use TSX-based templates instead of shitty text-based ones like Jinja.
    • Rust. Definitely the hardest option, but on the plus side you will learn Rust and that's a big plus.
  • source
  • parent
  • context
  • [–] 6 points 2 months ago (10 children)

    They can be guided. For example I asked Claude to benchmark some SystemVerilog.

    First it just did time ./run.sh and I was like "no, idiot. That's going to include the compilation/startup overhead."

    So then it said "I'll just increase the simulation time to amortise the startup cost." Again I told it that was bad and to actually measure the time in SV.

    So since SV has no wall-time functions, it used $system("date") or some bollocks like that. No! I said - use DPI-C.

    Finally it did the right thing. Would it have been quicker to do it myself? Absolutely not.

    Current AI is at lazy junior engineer level (but sped up 10000 times). You just have to make sure they don't make lazy dumb decisions. E.g. they're going to do everything in Bash if give half a chance.

    Tbf to AI I think they've learned a lot of this from humans. I bet if you searched for $system in SV you will find a lot of hand crafted horrors. I've seen human-written C code that found an ELF symbol by system("nm ... | grep ... | awk ..."). I shit you not.

  • source
  • parent
  • context
  • [–] 2 points 2 months ago

    Yeah... when the only documentation is CLAUDE.md...

    Still you can see why right? Improving docs/understandability for AI is good for them - they get more capable AI. Improving it for humans only helps other people, which is less of a priority.

  • source
  • parent
  • context
  • view more: ‹ prev next ›