🤖 NetBSD’s New Policy: No Place for AI-Created Code
— @linuxiac
“New development policy: code generated by a large language model or similar technology (e.g. ChatGPT, GitHub Copilot) is presumed to be tainted (i.e. of unclear copyright, not fitting NetBSD’s licensing goals) and cannot be committed to NetBSD.”
#NetBSD joins the ranks of software projects that ban #AI generated code.
How they are going to enforce such ban is an obvious question lingering in the air.
Does it include only cases like “hey #ChatGPT write a suite of unit tests for this class”? Or also cases where #Copilot simply autocompletes a for loop while I’m typing it?
In the latter case, how would a hypothetical reviewer enforce the ban? How would the for loop autocompleted by Copilot, or the boilerplate population of hashmap values, look any different than one I would write myself?
And if the issue is with any code that isn’t directly written by a human, then why stop at modern AI generation? Why not include LINTers and traditional IDE autocomplete features?
I have no doubt that the projects that are announcing these no-AI policies have good intentions, but it’s probably time for all of us to have an honest talk.
Code completion isn’t a clear cut binary feature. It’s a big spectrum that goes from the old exuberant ctags to ChatGPT writing whole classes.
And code completion shouldn’t be banned. If it makes a developer more productive, and if the developer understands the code that is being completed, then such bans are akin to a “drivers should only use cars with manual transmission because we feel that it’s more manly”. It’s a conservative and elitist act of shunning out new productive tools because we can’t understand them and regulate them properly.
And more people need to call out the bluff: in cases where the AI only completes a few lines of code, its basically impossible to tell if that snippet was written by a human or an AI assistant.
Ich bin gerade bei Microsoft Deutschland in Köln bei den Afterwork Prompts, einem Eventformat, das deutschlandweit stattfindet. Es sitzen hier Journalistïnnen von RTL und n-tv, freie, Deutschlandradio und mehr. Also ein Event für Journalisten.
Aktuell erklärt Alexandra Schroeder (Cloud Solution Architect bei Microsoft) wie Prompten funktioniert. #AI
Nächster Teil des Events bei Microsoft Deutschland: Uniper stellt vor, wie sie für alle Mitarbeiterïnnen den #Microsoft#Copilot eingeführt haben. Lernt man nebenbei noch was über die Gaswirtschaft. Es spricht Hans Petzold. https://www.linkedin.com/in/hanspezold#KI#AI
La nouvelle version de l’app mobile GitHub embarque le service « Copilot Chat », mais ne propose toujours pas le minimum de fonctionnalités pour gérer une organisation de plus de 300 dépôts. 🙄
Et la version web mobile est une plaie à utiliser.
It's a bit of a conflicting situation because it happens at the same time of my lab😁, but this session from @Jthake and @barnambora is one you don't want to miss! Get an overview of all the Copilot customization capabilities: plugins, connectors, custom copilots and many more! https://t.co/6pQy0FpBH9#copilot#microsoft365#msbuild
I still don’t understand the hype around GitHub #CoPilot.
Like lots of AI, it’s taking away the stuff that programmers generally enjoy (writing code) and giving us more of what we generally don’t enjoy (code review).
@mcc So developers will stop sharing information on #StackOverflow and future #Copilot and friends will be forever stuck in the past, answering questions about historically relevant frameworks and languages. #LLM#StuckOverflow
Habe den #Blogpost zum heutigen #StarWars-Day gerade erst freigeschaltet - und schon jetzt wird sie durch die Bing-#Copilot-KI präzise ausgewertet! 😵
Habe mal zwei KI-Antworten in die Drukos gesetzt und zögere gerade ernsthaft, ob ich nach der #Sturmtruppen-Theorie fragen soll. Ich wäre fast erleichtert, wenn die KI die Antwort noch nicht kennen würde...
Argh... reviewing a colleague's python code, and he's used ChatGPT or Copilot or some other AI "helper" to write a bunch of the code. And it's just got a load of "Why the hell have you done this?" bits. Especially in the unit tests - it really doesn't know how to write pytest tests.
If you don't know what you're doing it looks halfway reasonable, and it kind-of-sort-of works, but... you'd never write that code if you were doing it yourself.
I told #copilot I wanted an ASCII art of a volvo driving in Texas. It first told me I had to sign in to create images. I told it I wasn't asking for an image, but ASCII art. It then complied with something that doesn't look like a Volvo, and definitely has nothing that even stereotypically looks Texan.
"This is another variation of the High-Tech Illusion: the belief that software developers do easily automated work. Their principal work is human communication to organize the user's expressions of needs into formal procedure. That work will be necessary no matter how we change the life cycle."
With the {chattr} package, you can interact with GitHub Copilot and OpenAI's GPT models directly in RStudio.
• Chat with your models with a Shiny app that runs inside the RStudio IDE. Call it by running chattr_app()
• You can also use the chattr() function to send prompts to the LLM
I am extraordinarily annoyed by the fact that #Copilot appears to be enabled by default now in #VSCode and I can't find any way to turn it off. #dotnet#csharp