The test seems kind of dogshit, you could make the same argument against any tool, calculators or even abacuses would have the same effect.
I’m made to use it for work and it does speed up some tasks, however for some stuff it ends up being like the experiment where not doing the work the first time means the whole process takes longer at the end.
To add to this, we already know that context switching causes a loss in performance.
A person who’s thinking about how to solve a problem one way and then has to suddenly think about solving it in another way will perform worse.
The Neuroscience Behind the Pain
Context switching isn’t just annoying — it’s neurologically expensive. When you shift from debugging a race condition to answering emails, your brain doesn’t simply “change tabs.” It goes through a complex process:
-Memory consolidation: Storing your current mental model
-Attention disengagement: Breaking focus from the current task
-Cognitive reloading: Building a new mental model for the next task
-Re-engagement: Getting back into flow
Research from Carnegie Mellon shows that even brief interruptions can increase task completion time by up to 23%. For complex cognitive work like programming, this cost multiplies dramatically.
Here’s another article from CMU discussing the same thing: https://www.sei.cmu.edu/blog/addressing-the-detrimental-effects-of-context-switching-with-devops/
What this study shows is that a person who is faced with an unexpected context switch performs worse on a task than a user who has spent the last 12 questions performing the task the same way.
This exact problem would happen if you replaced AI with a calculator, or made a person swap from using paper to doing mental math. The problem here is context switching, not AI.
The way to ensure that the problem is AI and not the context switch, would be to continue the quest and see if the first group reverts back to baseline after 12 questions. 12 questions is how long the control group had to become acclimated to the task before their last context swap at the start of the test.
Also, of note, this is a paper on arXiv it is not published so it has not gone through a peer-review process which would certainly catch the failure to set a proper control group.
Context switching isn’t just X — it’s Y.
Are we sure this was written by a human?
AI being released was basically an apocalypse for people who use EM dash.
Here’s the most cited, human created (2001), paper on the topic of context switching performance loss: https://www.apa.org/pubs/journals/releases/xhp274763.pdf
Thanks.
And I’m all for em dashes. After all, I started using them after reading enough books. It’s just that particular construct that strikes me as especially LLM-y.
AI was trained on human writing. If it produces a certain tone, then that’s probably a result of the material that was favoured in training it. That construction was common in human writing before it became common in AI too.
What makes it stick out is when AI uses it in contexts where humans normally wouldn’t, but this kind of assertion is common in scientific papers and articles. It would make sense to train an AI on scientific writing, since that tone sounds authoritative and like you have some idea of what you’re talking about.
So I don’t think this is an LLM-construct; it’s an instance of the original style that LLMs copy.
True, but in my experience most people use a comma, not an em dash.
I’d like to see a study on that, I see it mentioned so much it’s almost achieved meme status.
It could very well be a Baader–(👀)Meinhof phenomenon.
That medium post is 100% LLM output.
100%, shit test
1980: TVs will fry your brain
1990: Videogames will fry your brain
2000: Computers will fry your brain
2010: Smartphones will fry your brain
2020: AI will fry your brain
Any takes for the 2030s?
Climate change.
Literally.
2030: Cyborg w/AI will fry your brain. Literally though.
Neural implants? Only this time they’re really going to fry your brain.
I mean, based fully on our current dystopian reality, I feel you just made a really good point about tech growing to a point where it fully captures you from reality, and indeed frys your brain by convincing you that fantasies are real.
MAGA is a great example of people with brains so fried they think a pedophile exconman with 34 felonies who killed millions of Americans trough a poor pandemic response is somehow helping them by destroying USAID, DEI, Healthcare, and Social Security.
Their brains are gonzo, all through the constant applied exploitation of all the tech you just mentioned combined.
AI will absolutley make it worse.
2030: Critical thought will fry your brain
Well looking around at where we are today, maybe TVs did fry our brains.
i think reading the title of this post hurt my brain. like what are we doing here? making medical claims using sensationalist and meaningless language… seems unhelpful
I think that if you use AI responsibly (as an assisting tool) like mentioned in the article, then you are pretty much on the safe side.
But when you have AI do everything for you, then there’s a big problem.
Personally I try not to use it at all, not a fan of all the problems that come with it.
AI is like a dog looking at itself in a mirror.
Some dogs are smart, and understand that this is a tool and that it is there to help you see things better… Some dogs are fucking morons and think their reflection is another dog, and they wanna fuck and fight…
There are a ton of good use cases for ai, and none of them include coquettish sexbots or drawings of me as a Simpson or a Ghibli sketch.
How do you know the dogs which want to fuck and fight aren’t the smarter ones?
What if the other dogs don’t recognize the reflection as anything meaningful — not a tool, a reflection, …? In that case, at least the “dumb” dogs figured out that something’s up.
Edit: anthropomorphizing the idea that nonchalant reactions = understanding well enough to not care. There’s many reasons any particular dog may not fight a mirror. Particularly, they may just rely less on vision to determine whether something is alive or not. That would not indicate understanding, though… it would indicate the dogs understandably passive approach to things which don’t seem to have any significance. Closer to a lack of awareness than an actual understanding of any kind.
I’ll never understand how an explosively imprecise, statistically luke-warm, grey goo extrusion sphincter could ever be mistaken for intelligence.
AI doesn’t exist, it’s a vacuous marketing term.
LLMs have vanishingly narrow legitimate, defensible use cases, but their output is intrinsically inaccurate, and should never be used without supervision from relevant domain experts.
@nonentity @technology I think the problem with your framing is it implies that humans are not also “explosively imprecise, statistically luke-warm, grey goo extrusion sphincter(s)”. We weren’t exactly living in a perfect world prior to AI, and all AI does is regurgitate what humans created. AI isn’t really changing the character of anything - and in several domains I’d argue it’s improving the baseline (coding for one).
It’s telling that you assumed the description applied exclusively to LLMs.
No one who persists in labelling LLMs as ‘AI’ should be treated as an authority on the subject, and I’d argue it’s one of the greatest indicators of how little they comprehend the situation.
THANK YOU! I studied AI in school, and it always bothers me when people think that LLMs are the only facet of AI. Between 2022-2024, I had a knee jerk reaction of explaining that AI is more than LLMs and that LLMs are really a small subset of the entire universe of AI, yadda yadda yadda. Now I’ve given up and roll my eyes as someone tries to tell me about the cool new Claude skill they built.
What’s funnier is people think I hate LLMs. That couldn’t be further from the truth; they are a fantastically interesting and innovative technology! “Attention is All You Need” is a great paper, and super impactful. I just hate that people are outsourcing their thinking to a chatbot and neglect the rest of my field of study.
I’m confused. Aren’t you the one who referred to LLMs In a thread that was conflating LLMs with AI? The parent’s comment seems to be right on point.
It’s kind of like how we’ve lost the war on hacking.
Large language models fall under the current definition of artificial intelligence just as much as Cyc or Cog did in their day, or various expert systems and machine learning models, diffusion models, etc.
Pretty much any non-deterministic inference engine can be classified as an AI, including LLMs.
theres plenty of legitimate use cases. your comment just sounds like youre repeating what everyone else says about it.
There could be many use cases, and some of them may even be legitimate, but I’m yet to observe any which have broad applicability, and they should only ever be wielded by a responsible, expert adult.
its a multitool, its applicability needent be broad imo. anyway, im glad we arent speaking in absolutes.
Oh boy wonder what kind of gatekeeping would come out of locking tech away unless you’re a “responsible expert adult”!
Jesus Christ not a single bit of thought went into what you said, did it?
You want to lock everyone but experts away? Who decides what an expert is, potus? The gop? The treasonous institute known as Stanford?
Is it only people with degrees? People making a certain amount? People that can pay a large licensing fee?
What a stupid thing to say.
‘AI’ fluffers sure do love the taste of grift flavoured tokens.
I’d ask what you were thinking, but it’s clear that played no material role in this extrusion. Extrapolating the assertion I constrained to a specific topic to the entirety of ‘tech’ is a bellowing straw man.
Further, the exclusively US centric examples of inappropriate stewards reveals the vantage to be squarely rooted inside that noxious bubble. The invocation of treason further betrays an affinity for national subservience.
To refine my original point, my observation of the application of LLMs is that the only entities who find them impressive are those who expressly lack proven expertise in the area it’s being applied. The correlation appears to be nearly linearly, inversely proportional.
LLMs could eventually prove innately useful, but there’s no indication they’re close to that, let alone traversing a relevant vector.
Personally, any world populated with entities who are impressed by LLMs is a world not worth living in.
I really do see the issue with AI. I see people around me outsource thinking to it too much. Like literally. As if they are happy that a machine can make their life choices for them. This is extremely worrying It’s About how people use it
I always thought recommendation algorithm will do it but the progress stopped at some point. We had apps recommending videos, music, feeds, news and so on for a long time but it never evolved into recommended careers or recommended places to live. Not in the sense where some algorithms that tracks you all the times tells you what your next important life choice should be. I don’t know anyone who’s using AI like that yet but I can see it happening in the future.
Thinking is hard and ppl would prefer to feel, instead. When you just have to vibe with your AI that thinks for you, ppl will absolutely use it and disempower themselves under the illusion of empowerment. They will infantilize themselves and end up being treated like the children they want to be.
I fucking hate this AI shit but I’ll admit I end up using Gemini (knowing its wrong sometimes) but it’s like how I’d use Google but just more of a complex ask instead of simple search query’s, I couldn’t imagine using it beyond that other than a follow-up or two.
It’s just a chatbot that has access to info, who goes onto their cable companies website and befriends the chatbot?
I have found Google search to be getting progressively worse where as I can type out a question to Gemini that will return better results than Google search. It’s annoying that Google search has gotten so bad and duckduckgo will return you something interesting but not relavent. So Gemini is my Google search nowadays.
It very well may be intentional; to drive people away from traditional search and in to Gemini.
Oh, do you mean Claudia!? She’s awesome!
Found the Richard Dawkins :P
I’ve used gpt a coup times when I was searching the web and forums for well over an hour and found nothing relevant enough to work. Theissue got solved in 5-10 minutes.
They enshittified the search so now using the chatbot is more useful. The search just returns slop and even fake slop forums.
I mean I have been using DDG for years now. I just could not find the right answer for my specific issue on my specific linux distro and AI was sadly just faster
Its more that SEO is so well known at this point you can whip up whatever AI generated garabge you want to be ranked high on search engines in seconds. For now the AIs are just better at “wading” through the trash since they somewhat curate the data its training on. Once all they can train it on is slop you better hope you still have some encyclopedias and text books laying around
Pretty much. Can’t find useful info without having to put in ALOT of extra work that I wouldn’t of a decade ago.
Fuck though I love being able to ask it for part numbers and info. Much less hassle to ask it then use the shitty corpo parts catalogues search features especially when there’s weird naming schemes and a lack of description, clicking through 50 parts trying to find the right one sucks.
I think the key point is that you’re not outsourcing critical thinking to LLMs, but are instead using it as a tool to do grunt work that you could’ve done yourself, but the LLM can pump out faster. This means constantly being critical of everything it does, asking questions, asking for links to credible sources, asking it to provide info to help evaluate the pros and cons of multiple approaches, with you making the decisions and learning along the way. Overall, any work a LLM produces that will have your name on it should be work you entirely understand and agree with. For coding, I find agent markdown files to be especially helpful to make sure the LLM follows my desired practices without me constantly making it refactor.
Largely, my assumption at this point is that LLMs may not always be around, so I definitely don’t want to be left holding the bag with a bunch of slop I can’t manage on my own. I think I’ll feel better when I can run open weight models on my own hardware that are fully competitive with cloud models. With models like Qwen 3.6 27B, it seems we are getting closer to that.
Those are important studies but nothing shocking. The conclusion to draw from them is the same one we’ve drawn from all technologies that have improved our lives to some degree: Without them, we tend to either be incompetent as losing access to them isn’t worth planning for, or we are demotivated because why would we deprive ourselves from technology that makes our work so much less exhausting?
It doesn’t necessarily remove our capacity to think (and the article falsely generalises to critical thinking), it shifts what kind of thinking we do.
If AI is as good or better than I am at writing code, then I’ll switch my brain to only do the orchestrating and architecture rather than the writing code part. And yes, if you remove AI, then the switch will cause me to perform less than I used to before AI, but not permanently, only until I get used to it again.
If an AI is better than a doctor at finding cancer indicators, then the doctor will focus their mind on finding solutions only rather than splitting it on both the detection and solution.
This is not new, not bad, and I’ll even go to the extent of saying it’s a great use of AI: Humans evolved for specialization. The less varied the tasks are, the better we are at the subset we specialize in. That’s what has driven our rapid technological and societal advances in the past millenia.
But, AI has many issues and many detrimental applications as well, so don’t see this comment as a full endorsement of AI.
Studies show that using a bulldozer for plowing a field decreases the farmers muscle density after just one day of use.
Christ. What a load of shit.
I don’t want it, all it does is to negate years of learned experience and ability to organically formulate ideas.
My experience with using ai, and at this point I’d say this experience is extensive / daily, is that it gets things wrong A LOT and with a high degree of confidence in its position.
In the early stages of using it I felt my problem solving desire start to slip, but after pushing through that and realizing I should not trust this any more than I’d trust human judgment it’s more like having another person to work with. That’s helpful but if I let me own thinking guard down at all I put myself in a lot of risk.
I hope most people that do use AI regularly eventually push through to this stage and we all will be way better off in the long run for the assistance.
I fear most people won’t push through. This study points to the obstacle, I’d love to see what can be done to help people overcome it, probably there’s room for AI usage training that we need to start to consider.
Should we trust a researcher whose brain got fried. Did they remember to do the old double-blind setup before the frying of the brains occurred?
Reaearchers: “Is the AI in the room with us now?”
Test Subjects: “No Asshole! You just took it from me while I was in the middle of using it!”
