Thought experiment: what if we _did_ teach intro CS with LLMs from the very beginning? Conventional wisdom is that this is a bad idea (students need to learn fundamentals, and LLMs can stand in for this), but I wonder if this is because we are thinking about it wrong. - ThreadSky

dbp.bsky.social • 150 days ago

Thought experiment: what if we _did_ teach intro CS with LLMs from the very beginning? Conventional wisdom is that this is a bad idea (students need to learn fundamentals, and LLMs can stand in for this), but I wonder if this is because we are thinking about it wrong.

Comments

guha-anderson.com•149 days ago

I’m concerned that this may put a ceiling on what one can achieve as a programmer. I’ll admit I am wrong when an exceptional freshman approaches me to do research and says they taught themselves to code with an LLM.

dbp.bsky.social•149 days ago

I agree with you to the extent that “taught themselves with an LLM” means using tools intended for people who know how to code (or to prevent them from needing to). But I wonder if we are still thinking too small (not sure if the ideas of my thread are useful, but they are _not_ just Copilot)

guha-anderson.com•149 days ago

I see. Idk. I think we should design CS1 by working backwards from what the rest of the curriculum needs. We are already were cramming way too much into the BS in CS. Now we have to also teach students to be better than ChatGPT at programming. It’s a lot to ask for.

dbp.bsky.social•150 days ago

e.g., experiments like https://github.com/copilotbook/CS1-LLM don't sit quite right with me -- giving copilot to beginners and saying lets not worry about the low-level details of programming.

dbp.bsky.social•150 days ago

Their results aren't bad (https://dl.acm.org/doi/10.1145/3649217.3653584), but I'm still pretty skeptical of this approach -- and whether it gives students the foundation to move on, etc. Learning happens when students are challenged (within reason!) and I'm skeptical if this sufficiently challenges them.

dbp.bsky.social•150 days ago

But back to the thought experiment. If we did want to do this, rather than just adding Copilot (which effectively thinks for students, nvm relying no them to have a sophisticated understanding of code to identify problems!), what if we used synthesis to alter the order in which we approach material.

dbp.bsky.social•150 days ago

i.e., start building programs by _only_: describing input / output data for problems, describing clearly how the input should be transformed to the output, and writing examples of how it should work. The code that is generated can then be validated, by hand or automatically, against those examples.

dbp.bsky.social•150 days ago

This process could continue with more sophisticated data, and we could explore problem decomposition, challenges in data representation, etc -- all before thinking about _code_ at all.

dbp.bsky.social•150 days ago

One of the challenges in teaching the design recipe (https://htdp.org/) is getting people to actually focus on those first three steps (signature, purpose, examples/tests), rather than just diving into code and getting themselves lost in the muck. Powerful synthesis might allow students to really get that.

Comments

Posting Rules

Reply