Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Building a no-code toxicity classifier by talking to GitHub Copilot

surgehq.ai

Post reply on HN
Print view View original

71–80 of 152 posts

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#71

Post by ShamelessC »

abeppu wrote: post #35

We're all focusing on the weaknesses of co-pilot (the comments can be longer than the code produced; you need to understand code to know when to elaborate your comment, etc). But also ... what do you need to know to recognize that the concept of a 'toxicity classifier' is likely broken? We can do _profanity_ detection pretty well, and without a huge amount of data. But with 1000 example comments, can you actually get…

> We're all focusing on the weaknesses of co-pilot (the comments can be longer than the code produced; you need to understand code to know when to elaborate your comment, etc).

See, this tells me you may not have even used copilot. Because while tutorials such as this (and the OpenAI codex tools) have you use comments explicitly to code, the reality is that you're not hammering out plain english requirements for copilot to work. You just code - and sometimes it finishes your thought, sometimes it doesn't. You hit tab to accept autocomplete, just like you would for any other autocomplete. So you are generally reading and evaluating what copilot thinks is a good output and choosing whether it goes in the program or not with the TAB key.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#72

Post by ShamelessC »

bmitc wrote: post #18

I don't really understand this. You're not coding directly in the language, but now you're coding in an implicit language provided by Copilot. From what I've seen on Copilot, although it is an impressive piece of tech, all it really points out is that code documentation and discovery is terrible. But I'm not for sure writing implicit code in comments is really a better approach than seeking ways to make discovery of…

You "came up" with the idea for intelligent autocomplete? And are you aware that this project actually required big innovations in language modeling and a supercomputer? Because I would say that is far more central to the concept behind the tech than the interface.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#73

Post by ShamelessC »

da39a3ee wrote: post #30
softwarebeware wrote: post #11

I'm out almost immediately. The first comment is more text than the code that it produces.

I think you're missing the point: this is starting to open the door to people who can't code.

I adore copilot and use it daily, but I'm pretty sure if I had always depended on it, I wouldn't be able to properly parse correct from incorrect programs.

It's a really really cool tool and a lot of these comments are just shallow dismissals from people who haven't actually used it and like to be reactionary on the internet because that's the world we live in apparently. But I think it works best when it's used by people with experience.

Hopefully future models with higher accuracy and research in grounding can get us to that point however.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#74

Post by michalhuman »

hartator wrote: post #5

Notice that the comments used to generate the code via GitHub Copilot are just another very inefficient programming language.

What makes it inefficient? It is verbose and similar to natural language. Given that code is more often read than written, isn't the code that is easier to understand more efficient?

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#75

Post by csomar »

hombre_fatal wrote: post #60

This is absolutely insane. I had no idea Copilot was this good. The negativity here just seems like sour grapes or weird goal posts. Sure, it makes mistakes and needs verification. But know what also makes mistakes and needs verification? All the code I already manually write as I tediously ratchet towards a solution. Removing some cycles from that process is a win. Just stubbing out close-enough boilerplate is a win…

Copilot is insanely brilliant and good. Its only issue is that it takes too little context (up to what your cursor is pointing on the file, at least on vim). If it did take all context (your whole project, maybe your shell history, your data files, the imported libraries code, GitHub repo issues/PRs, etc...) and it had some LSP checker for errors, add all of that to GPT-4; and maybe we'll have something that can do complex coding stuff auto-magically.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#76

Post by dusted »

This comment will of course be down voted, I'll attribute this to selection bias caused by the headline of the article.

You can't classify a comment as boolean toxic, toxicity does not exist in a vacuum. To extend the analogy from it's biological counterpart, toxicity depends on the organism. You should never just a piece of text in isolation and draw any conclusion about it. It must understood in context, both that of the subject, the recipient and the sender.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#77

Post by bmitc »

ShamelessC wrote: post #72
bmitc wrote: post #18

I don't really understand this. You're not coding directly in the language, but now you're coding in an implicit language provided by Copilot. From what I've seen on Copilot, although it is an impressive piece of tech, all it really points out is that code documentation and discovery is terrible. But I'm not for sure writing implicit code in comments is really a better approach than seeking ways to make discovery of…

You "came up" with the idea for intelligent autocomplete? And are you aware that this project actually required big innovations in language modeling and a supercomputer? Because I would say that is far more central to the concept behind the tech than the interface.

An idea is not an implementation, and I clearly mentioned it was an offhand comment in a casual conversation. My "idea" was exactly what I described above. Nothing more. I'm sure several had this idea, and Copilot was probably already in development. My comment was just a way to give a personal anecdote. I'm not sure what your point or complaint is. Did you somehow miss the reference to Office Space? It wasn't a serious claim. Just a segue to some thoughts I had.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#78

Post by ShamelessC »

bmitc wrote: post #77
ShamelessC wrote: post #72

Earlier quoted context omitted.

You "came up" with the idea for intelligent autocomplete? And are you aware that this project actually required big innovations in language modeling and a supercomputer? Because I would say that is far more central to the concept behind the tech than the interface.

An idea is not an implementation, and I clearly mentioned it was an offhand comment in a casual conversation. My "idea" was exactly what I described above. Nothing more. I'm sure several had this idea, and Copilot was probably already in development. My comment was just a way to give a personal anecdote. I'm not sure what your point or complaint is. Did you somehow miss the reference to Office Space? It wasn't a seri…

okay yeah - I apologize. In the context of other comments it seemed a little more dismissive of the tech itself. I see now that you were quite clearly going for humility. Should have caught it on the first read however, sorry again.

Re: Building a no-code toxicity classifier by talking to GitHub Copilot

#80

Post by chronolitus »

Reminds me of this post by Scott Aaronson: https://scottaaronson.blog/?p=6288

"Forget all that. Judged against where AI was 20-25 years ago, when I was a student, a dog is now holding meaningful conversations in English. And people are complaining that the dog isn’t a very eloquent orator, that it often makes grammatical errors and has to start again, that it took heroic effort to train it, and that it’s unclear how much the dog really understands."

Post reply on HN