Wow, you're saying that the thing with main selling point being approximate summarization of sources/text and guesstimation of solutions erodes skills related to parsing sources, thinking about text and comming up with solutions?

Who would have thought.

The more time goes on, the more I'm convinced that's a feature, not a bug.

While I think it's important to verify the things "we all know", I'm still filing this in the "No shit" section of the library.

Article is from june

Maybe this is a good thing if the SAT is the metric (partial sarcasm)

ChatGPT? what about claude?

the study: Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task

Some non peer reviewed paper with a tiny sample size is hardly enough to go "The Science has spoken". On the other hand, one can make an appeal to the intuitive idea that lack of practice implies lack of skill.

This immediately points to the pernicious effects of touting LLMs as a machine to do all the thinking for you. Heck, the enthusiasts are even using LLMs to do their social interactions for them.

This is a warning to people as much as workplaces: is this task you're offloading to LLMs really a skill you want to see atrophied?

And people brewing coffee on a keurig don't think about the process of making coffee as much as someone who grinds their own beans and brews a pour-over.

The output of the two is different. And while they can serve the same function (providing caffeine in the form of a drink), if someone needs artisan coffee, the keurig will not do.

And if someone has only ever used a keurig, they may not know where to start.

Anyways, I contend that we need to do a better job actually interesting people in artisan essays, so the slop produced by LLMs is more easily identified as such.

Of course that requires a desire to educate people for a reason other than extracting economic value from them, so it's unlikely to become widespread.

OK, let's slow down for one secound.

1. if you re-designed it with complex multiplication problem instead od writtng tasks, and calculators instead of chat gpt - and change nothing else - you would get exactly the same result. If you reduce the challange to minimum, how do you expect brain to respond?
2. The "we have this not-reviewed singular study based on 50 people we should raise the alarm NOW" is every sensational.

There are so many problems with AI, and we need so many checks and balances. Sure. But let's not change it into another "our pop-science vs their pop-science" kind of problem.

Ya THINK?