AI features enabled by default in education

[fonsp]

Should Pluto's (future) AI features be enabled by default for education?

Recap so far

See #3209 for previous discussion. Some points that came up in that issue (feel free to correct me if I misquoted you):

• @vchuravy "I would like to have exercise notebooks that don't use AI assistants, since I am trying to teach students how to debug software and write software" Option to change AI feature default to disabled #3209 (comment)

  • @fonsp "I'm also not sure about these features yet, they are currently in Pluto as an experiment." "I was also thinking of an option to disable AI features for a notebook file (something in the frontmatter), that you can put in your homework/lecture file. WDYT?"

• @pankgeorg "Let's discuss more about what responsibly using AI means, what we want to achieve and how we can get there." Option to change AI feature default to disabled #3209 (comment)

• @bvdmitri "If anything, I believe the use of new tools—AI-based or otherwise—should be encouraged, not discouraged. While these tools are still maturing, they clearly represent a shift in how we think about programming and developer productivity." Option to change AI feature default to disabled #3209 (comment)

And you?

What do you think? Feel free to add a comment! Also if this is your first time contributing to Pluto development!

[vchuravy]

I think there are two different perspectives that are useful to take into account.

The first is an educator with the goal of teaching people how to program and to teaching people the inner workings of computers, aka the “CS Professor”.

Secondly, an educator with the goal of teaching people a subject, let's say math, while using computers, aka the “Math Professor”.

For the “Math Professor” AI-based tooling for their students is perhaps a productivity win, since it reduces the support load that they have while using computers as a tool. Yet, we also must be careful that the AI tooling is not short-circuting the education goals, by creating the illusions in students minds that they understood a topic, while merely being in the passenger seat to an AI.

For the "CS Professor" this is doubly true, in the study of how to write, debug and analyze programs AI based tools often short-circuit the education goals and I want my students to "do it at least once manually" such that they build an understanding of how programs work fundamentally, and that might mean engaging with errors.

As an aside, for me #3224 is less objectionable since it is fully opt-in.

[bvdmitri]

I think it depends on what you mean by learning to program. Do you mean learning how to code, or how to build an app/service (a.k.a program)?

If the goal is to teach coding and debugging skills, then yes — AI tools can short-circuit that learning process. But if the goal is to help students build applications, then new tools are incredibly helpful. They remove barriers and let students focus on solving real problems.

For example, imagine a student with a great product idea but no interest in writing code by hand. Why should that stop them? Or why should they skip your course? They want to create a program after all. Well, now they are forced to learn how to code to express their idea. To code, they need to learn a whole new language — C, Julia, whatever. But why not use English? Julia is a language. English is a language too. Just with a different syntax rules. Now, we actually can use plain English, name it script.eng and let AI assistant (a compiler) to compile it to a program. Sure, now these "compilers" aren't very reliable, but its a question of time when they become reliable (or semi-reliable, as even current "true" compilers also have many bugs, e.g. Julia has 30 open issues labeled as "correctness bug").

Coding is just one possible tool to make programs and its perfectly fine and valid to have a course on coding. But in my view, teaching students how to build something meaningful is a bigger, more exciting goal than teaching them how to code line-by-line. If the aim of the Pluto project is to teach people to build apps than I would say AI assistance are paramount to that goal.

[vchuravy]

This is why I created two personalities. Your or how to build an app/service (a.k.a program)? is for me the "Math Professor" using computers to solve a problem, implement a service, doing something where programming is the means and not the goal.

Both goals are fine, but we should acknowledge that these are different goals that pose different requirements on the learning environment.

But why not use English?

Which is fine? But why then use Julia? There are plenty of zero-code environments out there?

If the aim of the Pluto project is to teach people to build apps than I would say AI assistance are paramount to that goal.

That seems quite far from what I understand the goals of the Pluto project to be.

[NOTtheMessiah]

There seem to be promising applications of LLMs, but I don't believe it's a good idea to make API calls without the users informed consent, it also presents a security risk if credentials or passwords are included in the code (and that can be incorporated within the training data). OpenAI in particular has had multiple controversies around training data. There should be options for local LLMs or services of choice, but it should never be forced upon the users.

Bad things happen when users do not understand or control how their data is being used:

https://techcrunch.com/2025/07/31/your-public-chatgpt-queries-are-getting-indexed-by-google-and-other-search-engines/

[fonsp]

We don't make API calls without informed consent 👍 Check out our docs: https://plutojl.org/en/docs/ai-editor-features/