19 comments

[ 3.4 ms ] story [ 61.2 ms ] thread
I love that this has finally come full-circle. I remember early scan-tron bubble sheets that you would record test answers on, then all the sheets were fed into a machine and it output the scores for the whole class.

Later on, there were web-based automatic graders, that, for example, would accept "3.0" as a correct answer, but reject "3" as incorrect.

This is just the natural recoil to a system that treats young human beings as automatons.

Used it to “write essays” and 100% as long as you give it a good look over, it’s incredibly difficult to tell it wasn’t written by a high schooler.
High school is actually going to be the worst intersection.

At the college level, an AI paper should still get you dinged for not understanding the material.

At the middle school/high school level, how the hell is a teacher supposed to tell the difference between a student who is learning to write and one who cheated with an AI?

Forgot kids cheating with text generators, many FAANG engineers got their position by using leetcode solving AIs, and many of these FAANG engineers are being promoted right now.

Blind users are unhappy of course.

>many FAANG engineers got their position by using leetcode solving AIs

What?

I feel like OP used an AI to generate a comment
I feel like most of the comments here are AI.
Actually, I'm starting to think: How useful can Github Copilot be to solve leetcode challenges? If you give it a comment like this:

// find the indexes of two elements of an array that sum to X

and a function header that was prompted to you by your leetcode engine, will it be able to fill it with correct code?

I mean: many people do Leetcode challenges for fun (or profit) and then publish their solutions on Github, so they were definitely part of Copilot training set.

I have Copilot and had coincidentally also been doing Leetcode just for fun, and yes, Copilot absolutely does give you the entire answer, if only because the questions are standardized and part of the GitHub training set.

Therefore, I turn off Copilot in any codebases I don't want to know the answers to, such as my Leetcode directory.

Disclaimer: I work for a university where I routinely route testing inputs to systems and people to make admissions decisions.

Teach the generator via an in-class markup of the output. Challenge the students to out-produce the AI.

There are testing systems that already use AI to grade essays via a first pass score before passing over to a human grader (who might be overworked). Better to open discuss the AI grader and how it works than act shocked that smart students will game the system.

Are they actually smart if they're bypassing the purpose of the challenges that are inherent to writing original work, that leads to them developing their mind?
My instruction to teach the generator. Basically:

1. Here is what the generator produces.

2. Here is why you will fail if you rely on it.

3. Assignment: Write a better essay using [these listed points] than the AI.

4. (extra credit) Write an AI that produces better output.

I put the burden on the educational system in terms of a challenge.

> 2. Here is why you will fail if you rely on it

Misses the point. Are students to use the generator should it not have the flaws you teach them? Is the purpose to deliver non-flawed essays, or to have students rise to the challenge of producing them?

Rising to the position of being able to use a good generator is not the goal.

Is the purpose of the challenge still relevant if you can achieve the same outcome with AI and some of editing?

I just feel this is yet another example of schools not being adapted to the Information Age.

Has anything really changed? It's not like paying a human to write your essay wasn't a thing.
Seems to match the common virtue/vice of tech: Now everyone can do it