“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition.
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog to summarize, not doing a thing that you never learned how to do won’t allow you to do it later.
-
@GossiTheDog From the "No Shit Sherlock" school of research
@ben @GossiTheDog Maybe so, but intuitive assertions have been catastrophically wrong before, so it's good to have hard data even for those.
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog OK, I guess I'll throw Anthropic a bone here and say it's pretty brave of them to post research that's fairly damning to their own products. I understand that's what they're "supposed" to do, but many orgs have a kill switch in upper management to suppress this sort of thing.
"Our results suggest that incorporating AI aggressively into the workplace, particularly with respect to software engineering, comes with trade-offs."
Ya think?!?!
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog If AI codes for you this is unsurprising since "you" are not really coding.
But i found AI very helpful when i want to know how to do something new. However you should be very critical about what you get back and poking and arguing with the answer you got can go a long way.
Not caring about the code you write is a more serious pitfall.
AI is a tremendous resource for anybody trying to learn.
-
@GossiTheDog I watched a thousand machine generated piano concerts, and I’m still not better at playing the piano. Strange.
@ahltorp @GossiTheDog well obviously once the player piano was invented, nobody needed to learn piano ever again.
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog almost like doing it is how you learn it?
-
@gabriel @GossiTheDog heavily depends on your prompt ask a simple slop question and not asking for explanations and trying to gain understanding and yes then absolutly true it can be used to cheat or it can be used to improve.
It's not about the quality of the answer. It's about allowing you to learn *how* to learn. Sometimes the best answer to a question is to ask "What do you think is the answer?" or "do you think there's a simpler way to go about figuring this out?". A LLM will always answer, instead of allowing you to access the creative part of your brain to find an answer for yourself. Develop creative/critical thinking. Etc. And you need creativity to build new things.
-
It's not about the quality of the answer. It's about allowing you to learn *how* to learn. Sometimes the best answer to a question is to ask "What do you think is the answer?" or "do you think there's a simpler way to go about figuring this out?". A LLM will always answer, instead of allowing you to access the creative part of your brain to find an answer for yourself. Develop creative/critical thinking. Etc. And you need creativity to build new things.
I've had mentees come to me for help figuring out something they were really close to figuring out for themselves. All I did was to say "You are close. What do you think you should do?". I was essentially a rubber duck. And they figured it out. All they needed was the confidence. If I would have answered, I would have taken away the dopamine they got after figuring it out themselves. And that is when learning happens.
-
N Marianne shared this topic
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog @vaurora in other news, someone found a tree in a forest.
-
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog Just like blindly following satnav reduces map reading and navigation skills. I am so glad I don’t have to use AI in a job context; I am a freelancer so can use it creatively, at least, I hope.
-
@GossiTheDog We’ll be busy removing this “digital asbestos” out of any software written this decade(?) for ages…
@fm_volker @GossiTheDog I mean, exactly. The asbestos comparison is one of the more apt I’ve seen. It’s being installed everywhere, promulgated as the best new thing for everyone, and ignoring all the experts who say “hey that’s a bad idea”.
Upside, those of us who understood mainframes and supercomputers were real sought after in y2k, and those that understand good data and information architecture will be the most valuable resource in five years.
Too bad most of us are old, and it’s gonna take a lot of capital to pull us out of mothballs to fix it all.
️ -
“Novice workers who rely heavily on AI to complete unfamiliar tasks may compromise their own skill acquisition. We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average.”
“We found that using AI assistance led to a statistically significant decrease in mastery.”
https://arxiv.org/pdf/2601.20245
https://www.anthropic.com/research/AI-assistance-coding-skills
@GossiTheDog Surprise of the century!!!1!1!1
-
@GossiTheDog this is no different than doing excessive hand holding as a mentor. When you answer questions instead of helping the mentee to fundamentally understand the thing they're asking about, and figuring out the issue for themselves (with your help), you take away from the valuable mental process that develops critical thinking and problem solving.
And a LLM does precisely this. It answers all your questions. You get the result. You never figure out the problem for yourself.
@gabriel @GossiTheDog what a great way to create dependency and increase subscription revenue (for Anthropic)
-
I've had mentees come to me for help figuring out something they were really close to figuring out for themselves. All I did was to say "You are close. What do you think you should do?". I was essentially a rubber duck. And they figured it out. All they needed was the confidence. If I would have answered, I would have taken away the dopamine they got after figuring it out themselves. And that is when learning happens.
@gabriel @GossiTheDog Is that how we learn today class we will learn physics . A 2 kg block sits on a frictionless horizontal surface. You apply a constant horizontal force of 10 N to the block for 3 seconds, then remove the force completely. What do you think the answer is?
-
R AodeRelay shared this topic
