@fasterthanlime@hachyderm.io titelbild
@fasterthanlime@hachyderm.io avatar

fasterthanlime

@fasterthanlime@hachyderm.io

hi, I'm amos! 🦀 I make articles & videos about how computers work 🐻‍❄ cool bear's less cool counterpart ✨ be kind

Dieses Profil is von einem föderierten Server und möglicherweise unvollständig. Auf der Original-Instanz anzeigen

fasterthanlime , an Random Englisch
@fasterthanlime@hachyderm.io avatar

Am on the go, left some work at home, remembered that 1) my home computer is on tailscale, 2) git can pull over SSH

(I would've pushed from an SSH session, but agent forwarding is broken in some way I can't debug)

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@0atman I was using it over a decade ago x)

fasterthanlime , an Random Englisch
@fasterthanlime@hachyderm.io avatar

asking LLMs to do the lord's work:

it rewrote it to 1 << 24

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

FAQ:

Q: Can't you do that translation off the top of your head???
A: No. Not anymore. I'm in my thirties now, leave me alone.

Q: Do you trust the LLM to get it right?
A: No, but it's trivially verifiable, and I do verify it.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@dngrs I get to do arrow-up to use the prompt on all the magic constants I find, so.

(also just for facts' sake, zed's inline assistant uses GPT-4o by default, not GitHub Copilot — completions are a different thing)

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@natty okay I forgot this frequently asked question:

Q: But this is trivial math
A: Yes but the assistant can do many other tasks I ask it to and it frees up my brain to think about other more important things

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@whitequark how much do you charge per token? 👼

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@natty yeah, I didn't even have to think about log2 before someone brought it up.

it's not like I don't know what logarithms are, it's just I'm wrestling a big bunch of context in a completely different area — the more I can delegate, the more focused I can be.

I know it sounds silly, believe me.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@natty I was already heavily using the assistant to help translate Go code to Rust (I do a lot of review + cleanups afterwards, and provide a lot of context to get quality results), I just thought "huh maybe I can clean up those magic numbers with it too" and I tried and it worked.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@dngrs Ohh I’m just now reading it correctly lol, that’s evil!

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@hisham_hm This is certainly one of the least impressive things I’m currently using LLMs for.

my perception of them has changed in the past couple years, and I’ve decided to accept that they’re here to stay. AI companies are overpromising, but LLMs are useful for certain things, and we better grapple with the labor implications sooner rather than later.

fasterthanlime , an Random Englisch
@fasterthanlime@hachyderm.io avatar

Not sure whether I can work with folks who don't believe that performance is a feature, moving forward.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

In fact, I would like to go and make a stronger claim: performance is enough of a differentiator.

Which means if the only thing difference between your product and the competitor’s is that yours is faster, then that’s good enough for me.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

The counter example here is some competitor to GitHub Copilot, which is much faster, but has much worse completions.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@sayrer I would honestly be very curious how mobile operating systems designed for performance would run like on those exact same chips android phones are sporting

fasterthanlime , an Random Englisch
@fasterthanlime@hachyderm.io avatar

I ran "golintci-lint" on that project I'm porting to Rust and I shouldn't have.

I shouldn't have.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@rakoo Although some linters in that collection are noisy and opinionated, many of them find real world problems before runtime. Try them on your projects and see if it finds some bugs!

fasterthanlime , an Random Englisch
@fasterthanlime@hachyderm.io avatar

Apple better announce June 10 that iOS 18 dictation is powered by Whisper..

Whatever thing they currently have is getting on my nerves 🙃

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar

@guilhermedea @Eramdam @ShadowJonathan @cadey I'm fairly sure DaVinci Resolve Studio's built-in speech to text uses Whisper, and it did fine with French on some test videos.

fasterthanlime OP ,
@fasterthanlime@hachyderm.io avatar
  • Alle
  • Abonniert
  • Moderiert
  • Favoriten
  • random
  • haupteingang
  • Alle Magazine