Skip navigation
Morning bathrobe rent: Training bias
0:00 / 5:22

Morning bathrobe rent: Training bias

Uncle Bob Martin

@unclebobmartin286K followers

Visit on X
236 likesDownloadOriginal post

33,232 views

Morning bathrobe rent: Training bias

21 Comments

Comments from the original post on X (Twitter)

  • thieme's profile picture

    @thieme

    I think the “AI produces crap” idea comes from NOT having put your constraints on them. Vibe coders aren’t able to do this. But AIs default to *not* have constraints, so by default they (currently) do “bad things”. In time they will default to amazing quality.

    7 likes

  • Parveen Yadav's profile picture

    @parveenyadv

    Is anyone else getting Heath Ledger’s Joker vibes listening to this

    4 likes

  • beersalt's profile picture

    @beersalt1

    It’s time Bob. Get on OnlyFans and let that bathrobe fly

    3 likes

  • Praveen B's profile picture

    @pben4ai

    Leftist Tears lofl

    2 likes

  • Jon Klaric's profile picture

    @complex_maths

    Let’s see if I can get the AIs to produce some clean and crisp FORTRAN77. Keep ‘em lean and treat ‘em mean to make the code clean!

    2 likes

  • Roman P's profile picture

    @RomanP918791

    Leftist tears 🤣🤣🤣

    2 likes

  • Andrew Rulnick's profile picture

    @MickeySteamboat

    PHP WASM! XD

    1 like

  • Hookshot's profile picture

    @3boatsdown

    Speaking of esoteric, I had mixed results when programming with BrightScript (roku), a fairly obscure and widely misunderstood language; it always ran but the code looked like crap. But that was about 6 months ago...

    1 like

  • af0r's profile picture

    @afOrtunado124

    such a blast to hear these, i feel like a kid in a campfire

    1 like

  • Tim Williams's profile picture

    @tim_willia98364

    For agents, RLHF makes a much larger difference than the training corpus alone. The better and more focused the RLHF post training, the better they are at individual languages. The 'how hard' a language is comes through the RLHF due to how hard it was for the trainers to work.

  • Alexey Lyakhov's profile picture

    @justfly1984

    Code quality is pretty average, with constraints it’s better, the question here is more about which languages has better constraints by default, and typescript is loosing to rust here 1:5

  • Jim Humelsine's profile picture

    @jhumelsine

    I was curious about APL. I asked ChatGPT to code the Game of Life. Here is part of it. It didn't execute in an online APL. Life ← { N ← +/ +⌿ ¯1 0 1 ∘.⌽ ¨ ¯1 0 1 ∘.⊖ ⊂⍵ N ← N - ⍵ ((N = 3) ∨ ⍵ ∧ N = 2) } I've forgotten too much APL myself to fix it:

  • William Rusnack's profile picture

    @WJRusnack

    I've been writing a lot of JavaScript and C lately it is much better at accomplishing what you want in javascript in my experience

  • whomever's profile picture

    @whatinthechain

    crazy old fart

  • Carovisi e agora?'s profile picture

    @carlosrvsilva

    Whats happened with the software craftsmanship? You're call yourself a software engineer now? Lol.

  • Uncle Bob Martin's profile picture

    @unclebobmartin

    The idea that software craftsmanship was ever about syntax juggling is horrifically naïve. The need for software craftsmanship has only gotten more intense. The agents will write vastly more code than humans can, and keeping that code clean is of supreme importance. We programmers must get out of the way as far as the syntax is concerned, and aggressively insert ourselves into the architecture and structure of the systems, right down to the level of individual functions.

    7 likes

  • Carovisi e agora?'s profile picture

    @carlosrvsilva

    So, I'm relieved you're still a crafstman!

  • Naoya's profile picture

    @AndrewNaoya

    You’re the best.

  • Benjamin Garcia's profile picture

    @ElbenjasSw

    Que es lo que trae esa thermo? LSD?

  • B's profile picture

    @CoolsieDev

    More importantly , do we have such skills as "crap analysis"?