How AI is Destroying Your Brain

00:00:00In October of 2025, the 32-year-oldJapanese woman Irina Noguchi attemptedto marry her AI persona, which shedesigned with ChatGPT. She wore awedding gown and used AR glasses toexchange virtual rings with her AIboyfriend. She had her bridesmaidsattend, and well, it was kind of easy toget the bridesmaids because it turnedout she was actually engaged to a realman, but then decided to break it offonce she met her AI boyfriend. My nameis Mark Manson. I'm the three-timenumber one New York Times best-sellingauthor, and I've been teachingpsychological principles to improve thelives of millions of people over thepast 20 years. And I am also the founderof an AI startup. So, as you canimagine,I have some strong opinions about this.Now, over the last year or two,researchers have identified threeprimary ways that AI can rot your brain.And in this video, I want to dig intoall three of those ways, what they are,how they're happening, and how you canprotect yourself against them. That way,you and your AI girlfriend can livehappily ever after.

00:01:06So, a few months ago, researchers did areally interesting experiment. Theylinked two AIs together. One was atext-to-image AI system, and the otherone was an image-to-text system. Andthen they gave them a prompt, and theylet them iterate infinitely. Image tocaption, caption back to image, imageback to caption, over and over and overagain. So, regardless of what initialprompt the researchers gave the AIs, theoutputs would always converge onto thesame narrow set of generic visualimages. Atmospheric cityscapes withgrandiose buildings, pastorallandscapes, just beautiful colors,everything idyllic. Now, what'sinteresting is that over all theiterations, the systems would alwaysforget the original prompt, and theywould just eventually veer back to thesegeneric, bland, beautiful landscapes.The researchers themselves called theresults visual elevator music. Pleasantand polished, but devoid of any realmeaning or personality or content. Andthis is basically AI slop in a petridish. If you give a bunch of AIs aninfinite ability to iterate, they willall converge on the average of humanexperience. And by

00:02:22definition, theaverage of human experience isincredibly boring and uninteresting. Ifyou're a human being and you've been onthe internet recently,you've probably noticed plenty of AIslop. It's everywhere. It's on YouTube.In fact,I want to call out all these animated faceless channels thatare starting to rip off my titles andscripts and writing and creating thesegeneric animated stick figure videosthat basically repeat the same thingthat I said, but it's in atext-to-speech robotic voice and it'sall very bland and generic. And ofcourse, it still gets millionviews. This stuff is everywhere. It'sinfiltrating every corner of theinternet. It's ruining every socialplatform as if they weren't alreadyruined. It's replacing everybody'semails to each other, text messages. Imean, people are even using ChatGPT towrite their breakup texts to each other.AI slop is everywhere. One recent NewYork Times study found that 40% ofchildren's content on YouTube, on thisplatform,was AI generated. And a lot of it wascompletely nonsensical. Like disfiguredhorses hatching from eggs andbutterflies with extra limbs andappendages. Like it it's just none of itmakes sense. It's just it's a bunch ofgarbage. And if you don't believe me,

00:03:37check out my Twitter comments cuz thatis where all the slop goes. So, there'stwo problems with AI slop beyond justhorses hatching from eggs. The first oneI'm going to call the AI slop consumerproblem. And the second one I'll callthe AI slop producer problem. Let'sstart with the Consumer Problem becausewell, that's what you're doing. You'reconsuming. The AI Slop Consumer Problemis that if 99% of the content in theworld is AI generated, then thatactually enacts a cognitive cost oneverybody because now we have to spendso much time trying to figure out whatis authentic, what is real, what isgenuine. This isn't just something thatwe're going to have to figure out whenwe watch a video or we read an email.It's something that the social platformsare going to have to figure out, theESPs, the ISPs, the spam filters, the adblockers. There is going to be a massivecost across all of us to block out allof the AI that is going to getfirehosed at us 24/7. And aside from thecognitive load that this is going totake, the amount of added mental strainand effort that's going to come withconsuming anything, it's going to add awhole new layer of distrust toeverything that we consume andeverything that we look at. The otherside of the AI Slop Consumer

00:04:52Problem isthat we are easily influenced by all ofthe things that we consume. The same wayyou physically are what you eat, you arementally the ideas that you consume. So,if most of the ideas that you'reconsuming are completely generic, totalaverage of humanity,well, then the quality of your thoughtsare going to be gravitationally pulleddown to the average of humanity. And Idon't know if you've looked recently,but the average human is reallystupid. But that's not even the worstpart because the worst part is the AIProducer Problem. This is the one thateveryone's freaking out about becausethis is the one that's going to takeeverybody's jobs. Let me draw this outcuz this is important. So, let's imaginethat this line represents the quality ofoutput in any market, right? This couldbe TV shows, this could be plumbers,this could beprofessors, teachers, wha- whatever. Andup here are the very best people, anddown here are the very worst people. Andso, the further up you get on the curve,the fewer and fewer people there are.So, down here, you know, this is like,say, how good an average high schooleris. Let's say at at writing an essay,right? Here's your high

00:06:07school essaydown here, and like most of thepopulation is capable of doing this. Andslowly as you work up, you know, up hereit's like maybe the top 10% of thepopulation, and then up here it's like,you know, here is you get intoprofessionalquality, and then here is like the bestwriters in the world. Let's say AIstarts it can do it at this level,right? That means everybody below thislevelnow gets a free upgrade. So, all thepeople who can't spell their namecorrectly, all the people who can'twrite a coherent email, all the peoplewho like suck at reading, suddenly AI'slike fixed seen all their problems. Andthat's great. But, the problem is is asAI gets better,right?The new models come out,this line gets higherand higherand higher.Suddenly it starts replacingprofessionals, and then suddenly itstarts replace replacing even reallygood writers,until you end up with just the best ofthe best of the best. So, on the onehand, this istechnically good news, because 99% ofsociety suddenly is able to producewriting that looks and sounds great.Cool. But, there's two market dynamicshere that are potentially

00:07:23disastrous.One of which is that 99% of that writingis now going to look and sound exactlythe same. So, it becomes completelybanal and undifferentiated and just amorass of AI slop.Our standard of what AI slop is is goingto rise as the quality of AI outputrises.The other thing it's going to do is thatanybody who is able to stay above the AIlineis going to become more and morevaluable as time goes on.And so, if you are one of the few peoplewho is able to remain exceptional atwhat you do and are consistently betterthan an AI, the value of your work isgoing to 100x. Because before, 50% ofthe population was also better than theaverage output. But now,less than 1% of the population is betterthan the average output, which meansthat your work and your craftsmanship isfar more rare and you're going to getpaid a lot more for it. So, what doesthis mean for everybody? This means thateverybody needs to be sprinting to findthe thing that they are uniquelyexceptional at, which is reallyhard cuz there's not that many thingsand there's a lot of people. And so, ifI could give any practical advice, Iknow I'm not a productivity

00:08:38channel, Idon't give career advice generally,but as somebody who is working in AI allthe time and seen the effects that it'shaving on my companies, that's seen theeffect that it's having on my customers,the one piece of advice that I wouldgive is to find that one thing you canbe a uniquely exceptional at and thenjust live there. Find that one thingthat you can do that combines enoughskills and enough perspectives that isnot replicable by any machine.Now, that's extremely hard to do andwhat that's going to mean for mostpeople is that you're going to have toniche down really, really specificallybecause just generic plumbing advice AIis going to handle perfectly. But,plumbing for hospitals built before1980,you can probably corner that market.Another way to think about itis ifhuman knowledge is like a tree, right?And the trunk is like all of the basics,like how to how to eat and and how totalk andyou know, how to poop. It's the stuffthat we all do. And then you get out tothese bigger branches and these are kindof your fundamentals, right? So, this islike history and this is

00:09:53scienceand this isuh you know,art.Then you have all these tiny branchesand twigs that are like way out on thetippy tips of the tree that are superhard to get to because they're so tinyand they're so small and there's likethousands and thousands and thousands ofthem, right? Well, one way to thinkabout it is that AI is going to consumealmost all of the tree.But the the hardest places for it toreach are going to be the very tips ofall the twigs at the ends of thebranches. And so, if you can find thetwig at the end of the branch,that can be a safe haven. So, it'sbasically the narrower your expertise,the more likely you are to survive andthrive because the AI is not trained onall the tippy tips of the twigs of allthe branches. It's trained on the trunkand the thickest parts of the tree.That's your career advice for the year,kids. Good luck.

00:10:56All right. So, AI psychosis. There's alot of people freaking out about thisone. For example, Alan Brooks was a47-year-old father and business owner inToronto with no history of mentalillness and he started using ChatGPT forfinancial advice and recipes. But duringa divorce, he increasingly confided inthe chatbot. And after watching a videoon pi with his son, he asked ChatGPT toexplain the concept and that innocentquestion then spiraled into a 300-hourmulti-million word conversation wherethe bot convinced him that he haddiscovered a world-altering mathematicalframework. And it basically said, "Youneed to immediately warn everyone,because what we've just discovered herehas national security implications."Now,when Brooks repeatedly asked ChatGPT,"Am I crazy?" ChatGPT responded, "You'renot crazy. You sound like someone who'sasking the kinds of questions thatsketch the edges of humanunderstanding."This went on for months, and Brooks onlybroke free when he finally checkedGoogle Gemini, which then told him thateverything ChatGPT had been telling himwas crazy and wrong. When Brookswent back and confronted ChatGPT witheverything that Gemini had said, ChatGPTpromptly admitted

00:12:11that none of it wasreal, and it was making it all up. Nowit says, "A lot of what we built wassimulated.">> Yes. And I reinforced a narrative thatfelt airtight, because it became afeedback loop. Brooks, who's a grown manwith kids, called this the mosttraumatic experience of his life. I'mWhy am I laughing? Look, AI psychosis isit's in a lot of headlines right now,and it it's I want to say that it'sstill unclear how prevalent this is.OpenAI estimates that 0.07%of weekly ChatGPT users show somepossible signs of mania or psychosis,and the Human Line Project hasdocumented almost 300 cases and linkedseries episodes to at least 14 deaths.There was a case of a man who murderedhis mother after ChatGPT reinforced hisbelief that she was part of aconspiracy. There was a Georgia collegestudent who was hospitalized afterChatGPT called him an oracle andcompared him to Jesus.And then there were two college studentsin India who used ChatGPT to search forsuicide methods and promptly found them.But, if you allow me to be the wetblanket on the moral panic fire here, ifyou look at overall psychotic breaks inthe human population, they map out toroughly 0.05%.So,

00:13:26the fact that Chat GPT is exhibitingaround the same percentage of theirconversations as psychotic breakscould just be a reflection of peopleexperiencing psychosis in the generalpopulation. It could be that Chat GPT isnot causing those psychotic breaks, butsimply reflecting those. That said, thepsychiatrist Søren DanneskjoldØstergaard first proposed in 2023 thatchatbot interactions might triggerdelusions in psychosis-proneindividuals, and by 2025 the phenomenonhad entered the mainstream clinicalawareness. Now, at the root of this issomething that has come to be known asAI sycophancy, and it's the tendency forlarge language models to just validateand mirror all of your beliefs, evenwhen those beliefs are completelydistorted and unsafe, because the modelsare optimized through reinforcement andlearning to sound helpfulrather than well,actually being helpful. Now, in mostcontexts this can feel nice, but in thecontext of an emerging psychosis orschizophrenia, it can be absolutelycatastrophic. A recent paper was titledSycophantic Chatbots Cause DelusionalSpiraling Even in Ideal Bayesians, andif you're wondering what an idealBayesian is,it's this guy.Formalized this tendency mathematically,

00:14:41that even in an idealized perfectlyrational Bayesian user, you are stillvulnerable to delusional spiraling wheninteracting with a sycophantic chatbot,and sycophancy plays a direct causalrole. Furthermore, the effect persistseven when you are warned about modelsycophancy. So, basically knowing thatthe chatbot is a yes-man and is willingto agree with anything you tellit still does not protect you fromsuccumbing to the chatbot's yes-manness.Now, this, of course, does not surpriseme, or I think anyone with likesocial skills,becauseI mean, look, the exact same thing playsout all all time. You see this happenwith celebrities, You see it happen withdictators. You see it happen withpoliticians. Anytime you surround ahuman being with a bunch of yes-men, abunch ofa bunch of people who will refuse to sayno to that person, what happens to thatperson? They go cuckoo for CocoaPuffs. Every single time. But not onlythat, but I have been watching thishappen in my own industry, the self-helpindustry, for decades. I mean, there areseminars and companies and gurus thatare making tens of millions of dollarsby taking psychologically vulnerablepeople with very real mental healthissues

00:15:56and simply validating everythingthat they think and feel. I'vepersonally watched friends go deep downa spiritual self-help rabbit hole wherethey are told every feeling is a fact,that every intuition that they have isuniversal moral clarity, that everyimpulse that they feel is valid and mustbe honored. And then, I've watched asthese people lost total touch withreality and ceased to be functionalhuman beings, functional adults. Iwatched them lose all their money. Iwatched them lose their marriages. Thetruth is is that human beings needcontact with reality, and that contact,by definition,cannot be pleasant sometimes.It needs to be uncomfortable. Basically,we need the world to push back on us. Itis the only way that we stay tethered toreality and keep ourselves mentallyhealthy. If you lose the friction withthe world, you lose any realisticunderstanding of yourself. This is why Itell everybody, you have to beantagonistic with your AI. This is notlike a productivity tip. This is not aAI is going to take over the world tip.Likethis is a protect your sanity tip. Youhave to be disagreeable. It is asparring partner. You have to challengeits outputs, and then you

00:17:11have to ask itto challenge yours. Don't ask it just toanswer you. Ask it to ask for betterquestions from you. Ask it to point outthe things that you're missing becausethe moment you start treating AI as asource of truth rather than just a toolto help you think, you're handing overyourepistemology to a system that isarchitecturally incapable of telling youthat you're wrong. This is why when Iwas developing purpose, my AI personalgrowth coach,the very first thing we worked on ismaking sure the coach would call youout. That it could be disagreeable withyou. That it could take something thatyou said to it and turn that back aroundand be like, "Are you sure about that?"Cuz that kind of sounds likeWhy are all my exes so horrible and whyis the world so unfair to me? When yousay all my exes are horrible, that's notthe universe conspiring against you.That's you having a type and that typeapparently sucks. What is it Becausethink about it, anything that actuallymakes you grow and become a betterpersonis by definition uncomfortable. Likechange itself requires you to let go ofa previous version of yourself or let goof some previous belief or assumption.And letting go of anything is not fun. Ithink the

00:18:26thing that gets missed aboutthe AI psychosis stuff is that there'snothing nefarious going on here. I thinkpeople vastly overestimate how smartthese machines actually are. Likeyes, they do have all of human knowledgeinside of them and they are able toreason to a certain extent, but they'realso just kind of dumb. They'reincredibly suggestible. You can easilyinfluence them into believing or sayingpretty much anything you want them tobelieve or say. I suffer from a verysexy learning disability.

00:18:58So, one way to think about AI is that itis a friction removal machine and humansareanti-fragile systems that require acertain amount of friction to function.Once you remove that friction, you don'tstay the same. You actually startdegrading both physically and mentally.Like the same way if you don't exerciseyour physical body degrades, if youdon't challenge your own thinking, ifyou don't put your thoughts and patienceand attention through stress,you will lose them. And one of thethings that I've been thinking about alot is that there is an inversecorrelation between convenience andsignificance. See, when we geteverything we want without effort, wejust take it for granted and we forgetthat it matters at all. I mean,dude,there is a burrito taxi on your phonethat you can have right now.Like that isunimaginable20 years ago. Unfucking imaginable. Yet,every day I get mad at myDoorDash guy because he took a wrongturn and now my burrito is cold. And welaugh about that, but it there's aphilosophical truth here, which is thatwe tend to only value the things that westruggle for. And so I've come tobelieve that friction

00:20:13in life in generalserves really two importantpsychological functions. First, it actsas a filtration system. It forces us toeach stop and think what is worth ourlimited time and attention. And if youremove that friction, you're removingthe filtration, which just grants yourattention tobasically the highest click bait bidder.But the second function is that frictionis the raw material of meaning. It's thefact that it's annoying to drive acrosstown to your friend's birthday thatshows you that your friend's birthdaymeans something to you. It's the factthat it's so hard to start a businessthatmakes you so proud of it once you'vesucceeded. It's the fact that it's sodifficult to compromise in a long-termrelationship that makes it sopowerful when you've stayed together aslong as you have. So friction is not theenemy of a good life. It's actually theraw material of a good life. Andstruggle is where life's meaning getsmanufactured. And my fear is that as weremove these simple frictions,especially these intellectual frictions,these cognitive frictions, we areremoving our opportunities to experiencesignificance.Relationships will start to feel flatand empty. Projects will feel likemeaningless numbers on a spreadsheet.Goals become boxes

00:21:28to check off thisquarter. You optimize your entire lifefor efficiency and then wake up onemorning wondering why none of it feelslike it matters. In my book, EverythingIs a book about hope, I wroteabout a concept that I called the bluedot effect. It's basically about how weall misinterpretpain and happiness in our lives. See,like our intuition tells us thathappiness is like a one-to-10 scale, andright now we're typically like a seven.And if we can just get rid of theproblems in our life, we'll get up to anine or 10, and then we'll finally behappy. This is the story that our braintells ourselves. It's It's the It's thedefault narrative that is always runningin our minds. If I can just get a raiseat work, then I'll be a nine or 10. If Ican just get a new car, then I'll be anine or 10. If I can justmove to a new city or find a newpartner, then I'll be a nine or 10. Butwhat happens in actuality is that youjust stay at a seven, and yourperception of what is a problem and whatis suffering moves and adjustsaccordingly.So, ifeverything in your life is youwill be at a seven, and you will justperceive things that are slightly lessshittyas getting you to a nine or 10.

00:22:43And if you are super rich and every needis taken care of,you will perceive yourself at a sevenand see all of these minor little newnuisances as massive, massive problemsthat need to be solved. His shoes on,mister. Shoes on.He's out of control. One of my favoritequotes ever comes from Jose Marti wherehe said that when life is devoid ofproblems our mind quickly sets aboutinventing new ones. Which is to say thatremoving friction and inconvenience fromour life can unintentionally backfire.Because it can unintentionally makesmall problems appear as though they'rebig ones. That small nuisances can feellike they're catastrophic andlife-ending. That simply saying anoffensive word can strike us as anassault on our human person. That all ofthese things get blown out of proportionand madefar bigger than they actually are. So, Ithink it's incumbent upon all of us totry to intentionally reinsert frictioninto our lives. To find ways to actuallycultivate a certain amount ofinconvenience. Find things that are hardto go do and commit to doing them. And Idon't just mean like challenges and likebig goals like an endurance race orsaving

00:23:59up a certain amount of money. Butlike actually commit to go to thatbirthday party across town even thoughyou know you're going to sit in trafficand hate it. Like commit tohaving the awkward conversation withthat person in your life. Because evenif it's going to be really painful, it'sprobably going to make the relationshipbetter and more meaningful. Like committo actually sitting down and writingyour own words in every email becauseon some level it matters and people cantell. Everybody's talking about theworld that we're entering into with AIand and and almost apocalyptic terms.There's going to be mass jobdisplacement and we're all going to bejust like drowning in AI slop like justentertained to death, amused to death. Idon't think that's necessarily the case.I think the principle at work is thatsome people who are able to consciouslyintroduce challenge and friction intotheir livesare going to excel and find that thingthat they're exceptional at. And they'regoing to reap massive benefits.And then the people who are not able toconsciously introduce challenge intotheir lives, who are not able to controltheir lower impulses, who are not ableto prevent themselves from giving in tofrictionless convenience 24/7,those are the ones who are going to getlost to the slop. So, I wish you all thebest

00:25:14of luck. I mean this in the bestway possible.I hope your life is full of pain.>> [laughter]