Thoughts on Safe Code Execution for Non-Developers

I really like Google is doing with it’s Gemini AI interface. Their “Canvas” mode surprised me.

It still has unintended tool calls (calling Canvas when it shouldn’t) but that can be fixed.

It evidently has pre-defined ways of generating a “page” as you chat, which is actually writing the same Python script with little changes requested by the user but it is useful for quick tasks such as being able to click a button and every time it makes a new image etc.

When inspecting the generated Python code, it is evident that it is tightly controlled by other modules also most probably written in Python by Google engineers.

This gives the whole experience a consistent flow and I liked that.

I still think Grok 3 and its interface (pre-determined software that encapsulates all the interactions with the LLM) surpasses its competitor (Gemini).

However the overall intelligence is very similar (except Gemini is a little more politically correct).

Gemini has the higher hand in stability though.

Back to the Canvas of Gemini, I liked the idea of importing React by default in a consistent approach to generate the interactions.

The Canvas is safe for non-developers to use but I still highly discourage non-technical people from copy-pasting random code and relying on it for sensitive tasks. Please don’t do that (unless you have a good understanding of how things work, at least at a fundamental but robust level)

It would be seriously irresponsible similar to how someone would learn to do surgery from YouTube and ask ChatGPT on the quick details and run into an open heart surgery. Same principle applies to critical software such as private data processing, costly API-powered AI tools, financial tools, and many more. You should never experiment with people’s sensitive data, including your own.

On the other hand, unlike some other developers, as long as you keep things “offline” I actually highly recommend anyone interested in algorithms, analytics, math and similar fields to use AI to experiment with code, let them generate code for you, and run things, tweak things, and be creative! 

But of course every creative action has its rules. Good thing is that, you can use AI tools to teach you these rules. Start with overall code safety, operating system safety, data safety and you are good to go. After these you will know what are to be left to professionals and what are safe to play around with. Good news is a lot of things are safe when you are offline!

If you are a Physicist, AI can generate you near-perfect 3D simulations of concepts you are working on by writing Python code in seconds. You don’t have to be a programmer. Just know how to properly write a prompt and describe your simulation in full detail.

I believe Gemini’s Canvas and other AI tools’ similar playgrounds will get better and safer over time, meaning every non-technical person will be able to safely execute AI-generated code right within the chat session! We are not too far from that!