ChatGPT On macOS Can Now Directly Edit Code

OpenAI's ChatGPT app for macOS now directly edits code in tools like Xcode, VS Code, and JetBrains. "Users can optionally turn on an 'auto-apply' mode so ChatGPT can make edits without the need for additional clicks," adds TechCrunch. The feature is available now for ChatGPT Plus, Pro, and Team users, and will expand to Enterprise, Edu, and free users next week. Windows support is coming "soon." From the report: Direct code editing builds on OpenAI's "work with apps" ChatGPT capability, which the company launched in beta in November 2024. "Work with apps" allows the ChatGPT app for macOS to read code in a handful of dev-focused coding environments, minimizing the need to copy and paste code into ChatGPT. With the ability to directly edit code, ChatGPT now competes more directly with popular AI coding tools like Cursor and GitHub Copilot. OpenAI reportedly has ambitions to launch a dedicated product to support software engineering in the months ahead.

Yep, but should it?

[gweihir]

I mean, seriously. I will have to fail sooo many people in my coding class this semester. Exam will probably be on paper. (Yes, not good. But better than the alternatives.)

Re: Yep, but should it?

[madsh]

I gave students a written test on object oriented programming on paper. Seemed like an easy way to grade the students actual understanding. And was super easy to mark and note and discuss with others.

Re:

[gweihir]

Oh, I will be fine. But many students will not. I would give them at least an IDE, but that institution does not have computer rooms anymore.

Re:

[mccalli]

Not all the world is a school. Can be handy to knock up examples or give starting points, I see no reason why this is a bad thing.

My own coding exams, in the 80s and very early 90s, were also on paper as well as a submitted project. It's not something I'd necessarily want to push everyone back to today. Algorithms in pseudo-code yep, but actual code? Should we also give them a bunch of cards to punch (which I also briefly coded in as pure academic exercise)?

Re:

[gweihir]

The problem is with those students that cannot resist the temptation of AI. I already, after they demoed the first exercise, has some strong candidates for the bets grades and some strong candidates for failure. Of course, the actual grading will be completely fair, but after having done this for a while you can tell. Before AI, some of those with low skills managed to notice in time and do something about it. That number is probably a lot smaller with AI.

Re:

[CalgaryD]

I can say that too

Here's the thing: AC is a human. He / she is not capable of research, reason, analysis.

The fact is that current chatbots operate somewhat similar to human brains, and remember a lot more textual information than an average human. So.. let us give them some benefit of the doubt.

Re:

[narcc]

The fact is that current chatbots operate somewhat similar to human brains, and remember a lot more textual information than an average human.

That is complete nonsense.

Re:

[CalgaryD]

What is it you disagree with? They (1) based on neural networks, the same building unit as biological brains, not exactly, but may be close enough, and (2) they use more text for training than any human reads in his life time. What part of it is nonsense?

Re:

[narcc]

They (1) based on neural networks, the same building unit as biological brains

That is complete nonsense. Artificial "neurons" are not similar to biological neurons in any way other than name. It's just a really, really, bad analogy that just so happens to also be really, really, good marketing. That bad analogy has lead many people, yourself included, to develop some seriously misguided ideas about what NNs are, how they work, and what they can do.

Re:

[CalgaryD]

Yes, sure, they are not the same as biological neurons. However, they may be functional enough to provide similar performance and similar behaviour. What makes you think that this is not true?

Research has been going into more realistic spiky neurons, however, we have not heard about significant advances nor any real advantages over the idealized CS neurons.

Sometimes 3.14 is close enough to the Pi number. Stating that it is not Pi, while correct, is often counter productive. And sometimes just plainly ri

Re:

[narcc]

What makes you think that this is not true?

It's very obviously not true. There is no similarity at all between biological neurons and the abstract things we inexplicably call 'artificial neurons' used in the equally inappropriately named 'neural networks'. They are nothing alike. It's just a bad analogy based on some overzealous guy's 1940's misunderstanding of brains.

That ridiculous name is only reason you, and many other laypersons, have this bizarre idea that NNs are somehow analogous to biological brains. They do not function the same way. T

Fab. About time too!

[seoras]

I was writing something new into an app yesterday and using ChatGPT wishing it would just apply it directly instead of having to cut'n'paste, especially for re-writes.

On Xcode, for example, programmatically adding constraints to views is now so easy with ChatGPT. It used to be a tiresome, time consuming chore.
Now just tell it how you'd like it to appear and it does the constraints, I only had to cut and paste the code.
With this update I can now tell it to code, then compile it and then tell it what I like/d

Been doing it for a while

[AlanObject]

My team has adopted Cursor (VScode fork) as the IDE of choice. (It isn't mandated but everyone likes it and we pay for a business account).

For quite a while you can do inline AI code editing and chat, and they have user-level and project-level rulebooks for you to set up and the AI to use. By default the LLM used is Claude, but you can select many others including OpenAI.

So I don't see why this is big news. And who uses Xcode anymore?

Thankfully

[rlwinm]

Thankfully my IDE is a mixture of vim and joe. Although I turn nonsense like color syntax highlighting off.