That is not what they found. What they found is that if you spend 60 minutes doing the homework with AI, you learn approximately as much as someone who spends 60 minutes doing it without AI. If you spend 20 minutes doing it with AI, you learn as much as someone who spends 20 minutes doing it without AI. Etc. The only real difference is that the AI people get way better homework scores, and as a result, tend to stop spending as much time, and thus tend to learn less.

That's what this preprint study found.

100% this. Even a short 5-minute quiz with just a couple questions at the start of each day will do the trick. But you simply can't rely on homework scores to be a proxy of learning or time spent studying anymore. Brief daily quizzes on the previous day's material will force them to.

Also, It's not like students have the excuse of, "I couldn't study, I didn't have anyone to explain X to me" anymore. The very AI that they might otherwise use to cheat on their homework can also explain to them the things they missed in last year's algebra class that are blocking them from learning this year's calculus, or whatnot. Congrats - you now have an infinitely patient personal tutor who will spend hours dumbing things down for you until you get it. Now you have no excuse - learn the material. It becomes all about actually putting the time in.

Yeah. There's been widespread misrepresentation of this (preprint!) in the media. What they actually found: test scores correspond to how much time you spend doing homework, regardless of whether you use AI or not. Which should not be a surprising result! Of course, the AI people, esp. those who had been using AI for more than a few months, stopped spending much time on their homework, and then their test scores fell to match the non-AI users who didn't spend much time in their homework either.

The TL/DR is, in today's age, you cannot use test scores as a proxy of how much time a user is studying or knows the material.

There are of course some caveats, beyond the fact that it's just a preprint - notably that there's nothing to separate why the user started using AI to do their homework, so there's potential confounders in there. For example, if a student is in a poor family, maybe they had to get a job to help pay the bills. But working a job itself can lower test scores - for example, having to work late the night before an exam. So there are some confounder problems with the study. That said, I'm sure the general conclusion is correct: time spent doing the thing = amount of learning the thing (relatively independent of what tools you're using in the process).

Actually Linux already does have this.

apt-get install nputop

Jokes aside it's worth noting this is distinct from GPU loads.

Define AI. Are you talking about having a heart to heart with ChatGPT? Or are you talking about AI in general?

AI is a huge field. Massive field. What am I using AI for? Well in no particular order:
- Improving the upscaled image quality of video games I play.
- Noise reductions due to missed rays cast in video games.

That's the normie stuff. Now to some serious things:
- A dedicated noise reduction system for removing CCD sensor noise from astronomy images without impacting the starfield (which against a dark sky is often indistinguishable from noise).
- I use it to predict compressor failure on oil platforms based on process conditions (we fed some 20 years of operational data in an AI module which correlated certain process conditions to compressor problems and built a predictive maintenance model for us).
- I use it for rapid storyboarding of training materials before I hand it off to a professional team to be drafted (I can't draw for shit).
- It is part of a recent update to simulation software which uses an AI algorithm to predict the most likely correct case for simulation cutting down simulation time from 8+ hours (I used to leave it run overnight) to a few 10s of minutes, and cut down the working set from TBs to a few 10s of GBs.
- We have an AI digital twin for multiple platforms which allow us to predict how control algorithms will affect production before we put it in practice. We used to do this in the past with a high definition simulator, but that took *years* to build for each facility. An AI model takes only a couple of weeks/months to build.

Oh and I also do boring shit with it:
- Translate between languages (it's MUCH better than old school translators).
- Prioritise email inbox after holidays (really good at picking out emails which contain due dates or urgent sounding instructions when you have 600 unread emails)
- Despite teams being the worlds smelliest turd, the AI summaries feature is pretty damn good for when you join a meeting late and want to know what you missed.

At home I used it to plan the design of my backyard since image generation was pretty good at creating good looking concepts. Nothing of construction quality but it did give me examples of what something may look like.

AI is vast. It does a lot of stuff. The things I mentioned are only things that one engineer in one industry is using it for. Love it or hate it, it's touching every field known to mankind, engineering, medicine, education, logistics, etc. The wife uses ChatGPT to generate math questions (not the actual math though, it sucks at that) but the concept of questions, e.g. "give me an exam question for pythagoras" and ChatGPT spits out something like "The sun is at a 45 degree angle and the empire state building is 2km tall, how long is the shadow cast?" And then she goes back and changes the question to say 443m tall, etc. But it's good for trying concepts.

GPUs were added to task manager a decade ago. We're talking about NPUs. They are completely separate hardware and hit the market about 3 years ago. When you have a GPU processing an AI workload it will show as a hit on the GPU's 3D engine use in task manager (as opposed to Copy, or the video encoders or decoders). If you have the NPU of an Intel Meteor lake chugging along, you have currently no insight into what is causing your processor to draw power. The only CPU load operations you see are the memory copy operations.

It's important to be able to see the load and what's loading your various processing components.

Well yes, but that doesn't mean that this particular case is in any way wrong or bad. Why wouldn't you want your task manager to be able to display the various workloads of different components in your computer? You like your computer fans blasting at full tilt without having a clue why and your task manager showing no meaningful load?

For the record Linux and BSD have done this since the beginning with various kernel modules for NPUs exposing a load level and tools released to monitor NPU load basically around the same time as NPUs hit the market. E.g. nputop basically was released the same time Intel released it's Meteorlake processors.

Like single engine? Single ECU? Your car is literally full of thousands of single points of failure with varying degrees of cost to replace them.

Does Slashdot not support the € symbol or are you doing something wrong?

Then you shouldn't want Android Auto or Car Play either. The question wasn't whether your wheels turn. The question was whether you should be demanding a car where functionality is tied to a device.

Thankfully we don't have such cars yet.

Why would you need that? 100% of cars on the market that use data in their dash support connecting to WiFI. You live in a world where you are tethered to your phone, so you are literally no worse off as your phone could provide connectivity. Meanwhile there are plenty of other people out in the world who want that full functionality without being dependent on the phone.

By the way what do you mean "deal with cloud syncing". Do you understand how "syncing" works? It's the exact opposite of you having to deal with anything. I get in my car and my Spotify playlist resumes from where my phone left off, my navigation from where Google Maps left off, just like how it would work if I plugged in my phone and used Android auto. Because that's... how cloud syncing works... I'm so confused about how you use technology. I don't know what you're doing but I feel pretty confident in saying you're doing it wrong.

That makes no sense. You just stated you swap devices attached to the car, that introduces compatibility risk over time. Will the new phone support the car? Probably, maybe not. For all my "intelligent connected" devices I demand 100% of the heavy lifting is done by the device with the longest lifetime. I expect my smart lights to work as lights with or without a phone. I expect my car to work with or without a phone. I expect my speakers to work with or without my phone (in fact this was a red line for me, I almost didn't buy them when I heard they came with an app, but was pushed over the edge when I heard you get 100% of that functionality by visiting the speakers IP address in your web browser).

Tethering any experience to a device that is shorter lived is stupid, even if we're not talking about a $50k expense.

No. Thank god the OP doesn't design cars. They were the ones saying they wanted such a thing and that car companies shouldn't develop systems of their own.

You don't? I recall enough that initially there were talks of banning them. There were multiple studies showing measurable increases in traffic accidents. We only put up with them because of the net positive they provide for navigation in general. In the end most places settled for regulations that were quite weak such as requiring manufacturers of devices and cars to display warnings about not using them while in motion.

Why do you think those warnings used to be littered across the screens back in the early days?

Who said "need"? Just because you use something doesn't mean you need to use it. I can add two large numbers together just fine, but I will be using a calculator since it's faster and easier to do so.

Also does a professor magically become perfect in all languages? You are commenting about a Bulgarian writing a paper in English, just so you know. Is being an English major part of the requirements for a math degree in a foreign country? Sounds silly if so.

I can see youve never used AI for anything. AI doesnt correct grammer or spelling. It restructuress phrases.

Now I typed that into Gemini asking it to correct it and along with pointing out the spelling mistakes it also said: "A more formal or academic revision of your statement: It is evident that you are unfamiliar with artificial intelligence applications. AI systems do not merely correct grammatical or orthographic errors; rather, they restructure syntactic phrasing. "

It's people like you who are making it detract from any point being made.

You're begging the question. Why do you think writing a paper is to elevate a skill? I've not written papers to elevate my skills since ... well... school. I write papers to convey information, mostly about the core skills and knowledge I do have. Whether the English is good or bad is irrelevant, just like it is irrelevant to the point the person was making.

Because the dumb student's goals as a student is to not be dumb obviously. Why do you think we teach people basic math before giving them calculators but are fine with having them use calculators later in life.

Were you home schooled, or did you just completely miss the point of what school is about?

LOL. Yeah sounds like slow internet is more than enough for your slow use case. That doesn't mean it's enough for everyone. Some people have some actual meaningful data requirements. Technology should not be paced to meet the needs of a person who doesn't use it. Jesus try playing Call of Duty. Not today mind you, it's certainly not going to download today on your slow arse connection. Maybe it'll be done in the morning if you leave your computer on overnight like we used to do back when Linux distros took 8 hours to download.

Was that fun?