LLMs often talk about ethics, responsibility, and fair use. However, during live Codeforces contests, the same models can still provide full problem solutions when users paste contest questions. This creates a clear contradiction.
LLMs are capable of recognizing Codeforces problems and detecting ongoing contests.
Given this, they could follow a simple rule:
Refuse to provide problem-specific solutions or code during live Codeforces contests.
It will not stop all cheating, but it can help limit the easiest form.
Open Question
Should LLMs refuse to help with live Codeforces contest problems? Would this be helpful or harmful?
Interested in hearing opinions from the community.
do you even know how machine learning models work before asking this question? say theres a very rare possiblity that you provide the question dataset to the model and it learns that you dont have to answer these questions. the questions language can be easily manipulated and itll answer that question
LLMs already enforce ethical restrictions in many areas. They refuse to explain how to make bombs, hack a system, or do illegal things—even though users try to manipulate questions to bypass filters but they still fails..
I think you misunderstood the blog. This has nothing to do with training models on new datasets. It’s purely about policy and prompt-time enforcement (e.g., detecting live contests and refusing problem-specific solutions), which LLMs already do in many other contexts.
you cant say you need to have a solution for finding least path between 2 cities and youre unable to write the code to it you can just come up with a different question saying i need least between 2 diff nodes? the llm will need to write djikstras and thats the solution for both questions.
why they should be? they are unconscious tools. Humans are bounded by ethics.
Ethics isn’t about the tool having feelings—it’s about humans deciding how the tool should behave.
So “LLMs are unconscious” isn’t an argument against ethical limits. It’s exactly why humans are responsible for setting them.
That's exactly what i meant.
You might find this blog post by Scott Aaronson interesting: https://scottaaronson.blog/?p=9333
tl;dr they don't care about students cheating when there are much more consequential abuses of LLMs that are actually worth worrying about.
LLMs being used for cheating isn't very high on the "things that would get us sued if we let them happen" list (like the current "remove clothing" scandal with grok) so nothing will be done about them by the companies.
just wait for the bubble to pop