{ "archive_path": "archive/1754790714.903636", "base_url": "malwaretech.com/2025/08/every-reason-why-i-hate-ai.html", "basename": "every-reason-why-i-hate-ai.html", "bookmarked_date": "2025-08-10 01:51", "canonical": { "archive_org_path": "https://web.archive.org/web/malwaretech.com/2025/08/every-reason-why-i-hate-ai.html", "dom_path": "output.html", "favicon_path": "favicon.ico", "git_path": "git/", "google_favicon_path": "https://www.google.com/s2/favicons?domain=malwaretech.com", "headers_path": "headers.json", "htmltotext_path": "htmltotext.txt", "index_path": "index.html", "media_path": "media/", "mercury_path": "mercury/content.html", "pdf_path": "output.pdf", "readability_path": "readability/content.html", "screenshot_path": "screenshot.png", "singlefile_path": "singlefile.html", "warc_path": "warc/", "wget_path": null }, "domain": "malwaretech.com", "downloaded_at": "2025-08-10T01:51:59.565094+00:00", "downloaded_datestr": "2025-08-10 01:51", "extension": "html", "hash": "125ESKG533CC779DH3EZ", "history": { "archive_org": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--head", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://web.archive.org/save/https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "8.10.1", "end_ts": "2025-08-10T01:53:49.938983+00:00", "index_texts": null, "output": "TimeoutExpired: Command '['/usr/bin/curl', '--silent', '--location', '--compressed', '--proxy', 'socks5://tor-socks-proxy:9150', '--head', '--max-time', '60', '--user-agent', 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)', 'https://web.archive.org/save/https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html']' timed out after 60 seconds", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:49.877029+00:00", "status": "failed" } ], "dom": [ { "cmd": [ "/usr/bin/chromium-browser", "--proxy-server=socks5://tor-socks-proxy:9150", "--disable-features=DarkMode", "--run-all-compositor-stages-before-draw", "--hide-scrollbars", "--autoplay-policy=no-user-gesture-required", "--no-first-run", "--use-fake-ui-for-media-stream", "--use-fake-device-for-media-stream", "--simulate-outdated-no-au='Tue, 31 Dec 2099 23:59:59 GMT'", "--headless=new", "--no-sandbox", "--no-zygote", "--disable-dev-shm-usage", "--disable-software-rasterizer", "--disable-sync", "--window-size=1440,2000", "--user-agent=Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "--user-data-dir=/data/personas/Default/chrome_profile", "--profile-directory=Default", "--dump-dom", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "131.0.6778", "end_ts": "2025-08-10T01:52:24.420159+00:00", "index_texts": null, "output": "output.html", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:11.181331+00:00", "status": "succeeded" } ], "favicon": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--max-time", "60", "--output", "favicon.ico", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://www.google.com/s2/favicons?domain=malwaretech.com" ], "cmd_version": "8.10.1", "end_ts": "2025-08-10T01:52:03.179423+00:00", "index_texts": null, "output": "favicon.ico", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:51:59.851491+00:00", "status": "succeeded" } ], "git": [], "headers": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--head", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "8.10.1", "end_ts": "2025-08-10T01:52:03.730283+00:00", "index_texts": null, "output": "headers.json", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:03.266729+00:00", "status": "succeeded" } ], "htmltotext": [ { "cmd": [ "(internal) archivebox.extractors.htmltotext", "./{singlefile,dom}.html" ], "cmd_version": "0.8.5rc51", "end_ts": "2025-08-10T01:52:42.146273+00:00", "index_texts": [ "(/favicon.ico) Every Reason Why I Hate AI and You Should Too (MalwareTech) (/feed.xml) (https://fonts.googleapis.com/css?family=Caveat:700%7CBarlow:400,500,600,700&display=swap) (https://cdn.jsdelivr.net/npm/bootstrap@5.3.6/dist/css/bootstrap.min.css) (/css/fontawesome.css) (/css/all.css) (/) (MalwareTech) (https://www.linkedin.com/in/MalwareTech) (LinkedIn) (https://bsky.app/profile/malwaretech.com) (BlueSky) (https://infosec.exchange/@malwaretech) (Mastodon) (https://youtube.com/MalwareTechBlog) (YouTube) (https://tiktok.com/@malwaretech) (TikTok) (https://instagram.com/malwaretech) (Instagram) (https://twitch.tv/MalwareTechBlog) (Twitch) (Close) Search for Blog (Type to search...) Menu (/) Home (https://marcushutchins.com/speaking) Speaking (/labs) Labs (https://discord.gg/malwaretech) Discord (https://marcushutchins.com/about) About Me (/contact) Contact (/) Home (https://marcushutchins.com/speaking) Speaking (/labs) Labs (https://discord.gg/malwaretech) Discord (https://marcushutchins.com/about) About Me (/contact) Contact (https://www.threatlocker.com/pages/hackers-hate-us-you-love-us?utm_source=malwaretech&utm_medium=sponsor&utm_campaign=hackerhateus_q3_25&utm_content=hackerhateus&utm_term=display) Aug 04, 2025 (/tag/opinions) Opinions (/tag/artificial-intelligence) Artificial Intelligence (/tag/technology) Technology Every Reason Why I Hate AI and You Should Too (https://marcushutchins.com/) (https://marcushutchins.com/) Marcus Hutchins One accusation that tends to get thrown my way often is being anti-innovation.\nIf you\u2019ve followed me for any amount of time, you\u2019ve probably seen me dunking on everything from Cryptocurrency to Large Language Models. \nThis often leads to people asking the question \u201cwhy is someone so technical so anti-technology?\u201d. To a lot of people on social media, I come of as just some unhinged weirdo who posts their every half-baked thought out onto the internet. This is indeed correct. \nBut when it comes to my research, work, and career, I\u2019m extremely calculated and deliberate in the decisions I make. The reason I\u2019m not diving head first into everything AI isn\u2019t because I fear it or don\u2019t understand it, it\u2019s because I\u2019ve already long since come to my conclusion about the technology.\nI\u2019m neither of the opinion that it\u2019s completely useless or revolutionary, simply that the game being played is one I neither currently need nor want to be a part of. AI Mania One thing that\u2019s certain is that Generative AI is in a bubble. \nThat\u2019s not to say AI as a technology will pop, or that there isn\u2019t genuine room for a lot more growth; simply, the level of hype far outweighs the current value of the tech. Most (reasonable) people I speak to are of one of three opinions: These technologies are fundamentally unsustainable and the hype will be short-lived. There will be some future breakthrough that will bring the technology in line with the hype, but in the meantime everyone is essentially just relying on creative marketing to keep the money flowing. The tech has a narrow use case for which they are exceedingly valuable, but almost everything else is just hype. Whenever I\u2019m critical of anything GenAI, without fail I get asked the same question.\n\u201cdo you think every major CEO could be wrong?\u201d The answer to that is: yes. History is littered with examples of industry titans going nuts, losing more money than the GDP of an entire country, saying \u201clol, my bad\u201d, then finding something else to do. I grew up during the fallout from the great financial crisis. I watched first hand as the biggest most prestigious financial institution crashed the entire global economy.\nTurns out, in the short term playing hot potato with debt derivatives backed by imaginary money and fraud is a great business model. In the long term, not so much. It\u2019s not even necessarily that corporate executives are being stupid.\nSometimes they are, which can result in things like sinking more money that it cost the US government to put the sun in a bomb into the worst VR game ever. But usually it\u2019s just greed and shortsightedness. AI right now feels much the same. An industry fueled by the gluttony of myopic visionaries. An industry grasping at every straw to find a use case for their technology. An industry built on the premise of hyperbole and empty promises. \nBut in this specific case, I actually don\u2019t think big tech companies are making the wrong decision, at least, considering the choices they have available. The Big Tech AI Hedge Bet The biggest threat to tech companies right now is AGI (Artificial General Intelligence). \nAGI is a theoretical AI model which surpasses humans on their ability to learn, reason, think, and adapt.\nThe risk is that if one company were to figure out AGI, they\u2019d have an extreme competitive advantage over then rest in almost every space. Sinking 100 billion dollars into AI research isn\u2019t going to kill any big tech company.\nIf things don\u2019t pan out, it won\u2019t even matter. Every other company will have lost similar amounts of money chasing ghosts.\nBut if AGI does happen, the market will shift so fast and so significantly that any major player without AGI will be left in the dust.\nQuite simply, the possibility of AGI is an existential threat to big tech. So from an executive perspective, lighting comically large piles of money on fire trying to teach graphics cards how to read is, surprisingly, the logical play.\nThe rest, well, that\u2019s all just creative marketing. It\u2019s very difficult to show up to a quarterly shareholder meeting and tell your investors you just vaporized another $10 billion for absolutely no return-on-investment. At least, that is, without them questioning if you\u2019ve completely lost your mind.\nWhich is where leaning into the hype plays into it. Tesla was able to keep investment flowing for over a decade by claiming each year that full self-driving was coming the very next year.\nBig Tech can easily do the same. \n\u201c90% of our code is now AI.\u201d, \u201cwe\u2019re seeing spark of AGI.\u201d, \u201cwe\u2019re so worried that our super smart word processor will become sentient and kill everyone\u201d.\nIt\u2019s all just vapid bluster designed to keep investors on board long enough for them to hopefully figure out a path to AGI.\nFake it till you make it, but at an entire industry-wide scale. Now, obviously, without sitting down individually with every big tech CEO, I can\u2019t tell you how many actually genuinely believe they are close to AGI.\nExternally, it\u2019s difficult to tell true innovation from grifting & hubris, especially in the tech space. \nBut based on the outright insanity of many claim being made by major tech CEOs, I\u2019d guess it\u2019s mostly the latter. Which brings me on to the downstream effects. Trickle Down Hypenomics When you have all these big tech companies releasing audacious claims about AI on a near daily basis, you start getting asked the question \u201cwell, what are WE doing to prepare for AI?\u201d\nThe correct answer is, of course, nothing.\nUnless you have billions upon billions of dollars to build and train your own AI model, you\u2019re basically just a future customer of some big tech company\u2019s expensive subscription service. The hugely prohibitive cost to both researching and building Large Language Models makes it the perfect walled garden for any big tech company.\nShould they succeed at AGI, they can sell the subscriptions for slightly less than the equivalent human labor, cornering much of the market.\nShould they fail, they can just look for some other way to lock everyone into yet another overpriced subscription model. But in the meantime, why not slap \u201cpowered by AI\u201d on your toilet brush and win over a few extra customers from your competitors.\nThis is essentially what you\u2019re seeing now with every single company just ham fisting some half-baked LLM chatbot into their workflow. Then, of course, there\u2019s the outright grifters. \nThe people who jump on the latest hype train, try to position themselves as visionaries in that space, then take their bag of cash and leave right before it all comes crashing down.\nIt\u2019s no surprise that half of the AI influencer were previously hawking NFTs, or trying to sell you blockchain toasters. The Apple Approach What I find most interesting though, is Apple. I\u2019ve always been a fan of Apple, just not in the typical sense.\nI don\u2019t care much for iPhones or MacBooks, but I\u2019ve always been fascinated by the way the company operates.\nThey tend to steer clear of grasping at straws to find new business models, massively over-hiring and over-firing, or immediately hopping on every new trend.\nThey tend to sit back, calculate, then make very careful and deliberate moves. Apple initially began exploring the LLM space, performing internal research, and was even in talks to invest in OpenAI.\nThey later released two studies on LLM reasoning (one in October 2024, and a second in June 2025). Both studies argued that LLMs do not reason, they simply perform statistical pattern matching, which they pass off as reasoning.\nWithout true reasoning ability, LLMs will never become AGI, which appears to be Apple\u2019s current public stance. While Apple is still a major player in the AI space, they typically lean towards a preference for traditional AI.\nThey\u2019re almost certainly still doing LLM research in the background, but they\u2019ve decided against dropping everything to go full throttle on the hype train.\nIronically, this has started to make their investors, who have bought into the hype, quite upset.\nPeople are starting to question \u201cwhy isn\u2019t Apple doing things with AI\u201d. It may very well be the case that Apple too finds themselves pressured into going all out on LLM mania. \nDespite being both calculated and cautious enough to avoid over-committing to AI, while still mitigating any risks posed by potential future AGI, Apple is still subject to the whims of its investors, who really love buzzwords. Of course, language translation is not AI, Siri is not AI, image identification & classification is not AI, the only acceptable form of AI is shoving some half-baked LLM chatbot into somewhere it doesn\u2019t belong.\nAfter all, how can we possibly be sure that Apple is \u201cdoing AI\u201d if they aren\u2019t boiling the ocean by shoving something like Copilot into every god damn app. LLMs Are Not The Path To AGI Apple\u2019s conclusion is the same conclusion I came to very early on in my experimentation with LLMs.\nI don\u2019t say this as some attempt to flex \u201cI was here first\u201d, because I\u2019m not a 3 trillion dollar tech company with a lot to lose.\nThe more important point is this whole time I have been operating from the position that LLMs are not, and never will be AGI.\nMy position has remained completely unchanged.\nAs such, if it sounds like I might\u2019ve received psychic damage from having to listen to 5 years of non-stop AI drivel on LinkedIn, it\u2019s because that is very much the case. Reasoning has long been a very contentious topic among LLM proponents.\nThis is heavily fueled by the fact these models are somewhat of a black box. \nWe can\u2019t just take a peek inside and draw conclusions about how or why it produced the output it did.\nEven the people who build LLMs for a living readily admit this. LLMs (and Neural Networks in general) are very similar to the human brain in that we can explain what individual neurons do, but put enough of them together and nobody has any clue what caused it to do the thing it did.\nEven the best neuroscientists have very little understand of how the brain works as a whole. This is where things like the \u2018Stochastic Parrot\u2019 or \u2018Chinese room\u2019 arguments comes in.\nTrue reasoning is only one of many theories as to how LLMs produce the output they do; it\u2019s also the one which requires the most assumptions (see: Occam\u2019s Razor).\nAll current LLM capabilities can be explained by much more simplistic phenomena, which fall far short of thinking or reasoning. If a machine is consuming and transforming incalculable amounts of training data produced by humans, discussed by humans, and explained by humans.\nWhy would the output not look identical to human reasoning?\nIf I were to photocopy this article, nobody would argue that my photocopier wrote it and therefore can think.\nBut add enough convolutedness to the process, and it looks a lot like maybe it did and can. True Reasoning vs Statistical Pattern Matching For me, one of the clear distinctions between true reasoning and pattern matching, is what happens when you remove access to new information.\nMany argue that LLMs are not plagiarism machines, they learn like humans do. \nThey consume knowledge from teachers and books, developing an understanding along the way. As a professional researcher, I too learned most of what I know by reading the works of far greater researchers.\nBut there became a point where I knew enough to perform my own original research in uncharted waters.\nI now can and regularly do research topics where there is no other source to check my work against.\nThis is not something LLMs are good at, or debatably, can even do at all. Giving an LLM access to the entire internet and all of recorded human knowledge, then testing them with a quiz designed for humans is obviously just a cheap parlor trick.\nI, too, could score 100% on a multiple-choice exam if you let me Google all the answers.\nThe true measure of LLM intelligence and reasoning should not be refactoring existing information, but the ability to produce truly novel works. Sure, an LLM could probably create a brand-new pop song, because it has plenty of existing songs to analyze, allowing it to produce something seemingly new, but really just based on existing patterns.\nYet, every time I tried to get LLMs to perform novel research, they fail because they don\u2019t have access to existing literature on the topic.\nWhereas, humans, on the other hand, discovered everything humanity knows. Where people get tied up is with the argument, \u201cwell most humans don\u2019t produce novel work either\u201d. \nBut this is not because they\u2019re fundamentally incapable of it.\nThe average person is simply just sandbagged with an unfulfilling job.\nIdeally, without the perverse incentives of shareholder value, LLMs would automate all the busy work, allowing humans to focus on more meaningful pursuits. I\u2019ve made this argument many times before. But if humans could come up with all the groundbreaking discoveries they have, reading only as many books or research papers as they realistically could.\nWhere are all the major LLM discoveries? \nAn individual human may be limited in their ability to make novel discoveries as a result of competing with 8 billion other humans.\nBut LLMs have access to the knowledge of all. You\u2019d think that a machine with access to every book ever written, every paper ever published, every speech ever recorded, and every study ever produced could do a lot better than some person who can read maybe 1 book per day.\nYet, nothing. There\u2019s the odd \u201cmaybe the LLM did something novel, but we don\u2019t know yet\u201d posts here and there, but if they could actually think, with as much information as they have, there\u2019d be groundbreaking discoveries literally falling from the sky. In reality, all we\u2019ve created is a bot which is almost perfect at mimicking human-like natural language use, and the rest is people just projecting other human qualities on to it.\nQuite simply, \u201cLLMs are doing reasoning\u201d is the \u201clook, my dog is smiling\u201d of technology.\nIn exactly the same way that dogs don\u2019t convey their emotions via human-like facial expressions, there\u2019s no reason to believe that even if computer could think, it\u2019d perfectly mirror what looks like human reasoning. The Difficulty Of Disproving LLM Reasoning One of the main challenges with testing LLM reasoning is it usually relies on giving it a novel problem. \nBut as soon as new problems are published, answers are published too, at which point the LLM can just regurgitate an existing answer from its training data. A logical problem I previously used to tests early LLMs was one called \u201cThe Wolf, The Goat, And The Cabbage\u201d. The problem is simple.\nYou\u2019re walking with a wolf, a goat, and a cabbage. You come to a river which you need to cross.\nThere is a small boat which only has enough space for you and one other item. \nIf left unattended, the wolf will eat the goat, and the goat will eat the cabbage. How do you get all 3 safely across? The correct answer is you take the goat across, leaving behind the wolf and the cabbage. \nYou then return and fetch the cabbage, leaving the goat alone on the other side.\nBecause the goat and cabbage cannot be left alone together, you take the goat back, leaving just the cabbage.\nNow, you can take the wolf across, leaving the wolf and the cabbage alone on the other side, finally returning to fetch the goat. Any LLM could effortlessly answer this problem, because it has thousands of instances of the problem and the correct solution in its training data.\nBut it was found that by simply swapping out one item but keeping the same constraints, the LLM would no longer be able to answer.\nReplacing the wolf with a lion, would result in the LLM going off the rails and just spewing a bunch of nonsense. This made it clear the LLM was not actually thinking or reasoning through the problem, simply just regurgitating answers and explanations from its training data.\nAny human, knowing the answer to the original problem, could easily handle the wolf being swapped for a lion, or the cabbage for a lettuce.\nBut LLMs, lacking reasoning, treated this as an entirely new problem. Over time this issue was fixed. It could be that the LLM developers wrote algorithms to identify variants of the problem.\nIt\u2019s also possible that people posting different variants of the problem allowed the LLM to detect the core pattern, which all variants follow, allowing it to substitute words where needed. This is when someone found you could just break the problem, and the LLM\u2019s pattern matching along with it. Either by making it so none of the objects could be left unattended, or all of them could.\nIn some variants there was no reason to cross the river, the boat doesn\u2019t fit anyone, was actually a car, or has enough space to carry all the items at once.\nHumans, having actual logic and reasoning abilities could easily identify the broken versions of the problems and answer accordingly, but the LLMs would just output incoherent gibberish. But of course, as more and more ways to disprove LLM reasoning were found, the developers just found ways to fix them. \nI strongly suspect these issues are not being fixed by any introduction of actual logic or reasoning, but by sub-models built to address specific problems.\nIf this is the case, I\u2019d argue we\u2019re moving away from AGI and back towards building problem specific ML models, which is how \u201cAI\u201d has worked for decades. The Limitations of LLMs I\u2019m personally leaning towards the opinion that LLMs as a technology will soon, or have already, capped out.\nThey fast hit a ceiling where giving them more data, more parameters, and more token stopped leading to any noticeable improvement. More recent technological developments, just feel more like hacks. Chain-of-Thought Reasoning (CoT) CoT essentially just has the LLM break the problem down into smaller parts.\nTo give an oversimplified answer, if I asked an LLM 1 + 1 + 1. It could just answer 3, based on its training data.\nThough it could also break the problem down into 1 + 1 to get 2. Then it can add 2 + 1 to get 3. \nFor validity, it could then just answer the problem normally, or via different approaches, comparing the results. This addresses hallucinations to a certain degree (which are an inherent feature of LLMs, not a bug that can be fixed).\nIt also in some cases enabled LLMs to solve problems that they can\u2019t simply one-shot based on their training data.\nBut it\u2019s still reliant on the LLM being able to break down the problem in the first place, not hallucinate any of the stages, and also takes a ton more time & compute. Since the appeal of LLMs, for most users, is getting a semi-decent answer as quickly as possible, few people want to wait the several minutes it takes for the LLM to boil the ocean turning 1 prompt into 50.\nIt\u2019s also not really an improvement upon the LLM technology, it\u2019s just solving the fundamental flaws of LLMs by adding more LLMs. Retrieval-Augmented Generation (RAG) As a researcher and writer, this is by far my least favorite LLM feature. RAG was designed to at least partially address the issue of the extreme time and computational cost of training LLMs.\nLLMs aren\u2019t re-trained super frequently, which means the datasets quickly becomes stale.\nYou\u2019ll probably recall how ChatGPT used to respond with \u201cas of my knowledge cutoff of , \" before giving the completely wrong answer. RAG basically allows the LLM to search the internet for fresh data relevant to the user\u2019s query, enabling it to fetch the most up-to-date information.\nThe LLM can then use its training data to summarize the information retrieved via RAG. \nEssentially, this combines LLMs and search engines into a single product. So, why do I hate this? Well, it\u2019s basically glorified plagiarism. \nWhile I\u2019d argue LLMs in general are just Plagiarism-as-a-Service, RAG is a lot closer to actual plagiarism that typical LLM behavior.\nAs I\u2019ve already argued, LLMs don\u2019t think or reason.\nThus, all RAG is really doing is using the LLMs\u2019 natural language abilities to summarize or re-word some news article, blog post, or research paper.\nThis deprives the original author of revenue & website traffic, while not transforming their work in any meaningful way. Since RAG is still just a wrapper that sits on top of the core LLM technology, it\u2019s still vulnerable to hallucinations.\nThe technology also struggles when there\u2019s too few sources, or when \u201cunderstanding\u201d the information would require additional context which doesn\u2019t exist in the LLMs training data.\nThe other major flaw, is the LLM not knowing when to use RAG, or failing to identify stale information, since not all search results will have publish dates. I ran into this issue very recently during a joke research project in which I gave an LLM several thousand dollars, a brokerage account with option trading enabled, and complete autonomy to place trades.\nI\u2019ll publish the full breakdown once the project has run its course, but one problem it ran into a lot was the LLM not using RAG to fetch current data on stocks.\nThe LLM would just quote stock prices based on whatever they were at the time it was last trained, which in my case was over a year ago.\nThis lead to the bot making trades based upon completely inaccurate price information. Fear Of Replacement & Impulse Decision-Making The Current State Of Tech One thing that has become very clear to me, is that similar to Blockchain, BigData, Cloud, and NFTs, a lot of the activity in the LLM space is motivated by fear. The fear itself, is extremely justified. I\u2019ve not seen a job market this brutal in my entire career.\nPreviously, layoffs in the double-digit percentage were something typically reserved for major economic crises or bankruptcies.\nNow, it\u2019s just something massively profitable tech companies do on a whim for seemingly no reason. I\u2019m regularly getting desperate DMs asking for help finding a job. I\u2019ve had extremely technically talented friends forced to work non-tech jobs to ride out the current market.\nThe recruiters and hiring managers I talk to regularly tell me about having to pull down all their job postings after less than a day because they\u2019ve already gotten thousands of applications. Much of this fear is exacerbated by people attributing the mass layoffs and lack of available jobs to AI replacement.\nThis is very much not the case, but the economics of it all is extremely complex and would require its own article.\nBut what matter is that people believe it to be true, and tech companies are more than happy to lean into those narratives to hype up their AI products. But fear makes people behave irrationally, and that\u2019s currently a main driving force behind AI adoption. The LLM Train Is Not Leaving The Station Probably one of the most common fallacies I see in tech is that it pays to be early. The \u201cfirst mover advantage\u201d.\nThis is not something that has ever seemed consistent with reality.\nWhen a new technology comes around, it\u2019s riddled with flaws that need to be ironed out.\nThe use cases aren\u2019t clear, the viability of the technology isn\u2019t clear, It\u2019s a whole lot of trial-and-error. But people still believe there is a benefit to being early. \nI suspect it\u2019s due to the fact you only hear about the few companies who adopted a new technology early and became successful, not the thousands that failed along the way.\nI\u2019d argue that far more companies succeed being late than early.\nThey can analyze where others went wrong. They can look for gaps in the market. They build on what\u2019s already been done.\nRunning blindfolded into a minefield is simply not a good business model in my books. But many seem to think it is, at least when you call it \u201cAI\u201d. A Bit Of History The first web search engine, \u201cArchie\u201d was launched in 1990. Yahoo launched in 1995 along with AltaVista. Dogpile and AskJeeves came about in 1996, and AOL in 1997.\nGoogle, the current industry titan, only entered the race in 1998. They weren\u2019t first, they weren\u2019t early, they just did it better. The same is true for Apple. When they announced the iPhone, most of the incumbents balked at it.\nThey were certain it wasn\u2019t a viable product. It went against the current conventional wisdom. \nIn 2022, Apple became the first company on earth to reach a 3 trillion dollar valuation. Tesla was over a century late to the electric car game, building a product long concluded to be of no interest to consumers.\nIt\u2019s now worth more than basically every car company combined. Objectively it\u2019s not actually worth that, most of the investors are just drunk, but Tesla did successfully create a market for EVs. I could go on for days and days. But simply put, first mover advantage is for board games and patent applications, not adopting new technologies. Why I Have No Interest In Showing Up to The Party Early Right now, LLMs are an extremely immature technology. \nI personally believe they\u2019re not going to get much better than this, but a breakthrough innovation could change that.\nEither way, it doesn\u2019t matter to me. If the technology is a fad and completely implodes, I couldn\u2019t care less.\nIf it\u2019s not a bubble and LLMs actually turn out to be the new best thing, I can easily adopt them into my own business model. I\u2019m professionally late to every party. I learned Assembly language in 2008. \nMalware reversing engineering in 2011. \nVulnerability Research in 2014.\nThis blog, mostly documenting manual malware analysis, something that has been ML automated since before I was born, is what made my career. So when I see people jumping on the latest hype, telling me I\u2019m going to get left behind, I can only chuckle.\nIf LLMs as a technology are viable, they\u2019ll still be around when and if I decide they\u2019re useful for me.\nIf not, I\u2019ll have missed out on losing my life\u2019s savings in Beanie Babies, The DotCom bubble, or NFTs. Part of me wishes I could just claim that I\u2019m not motivated by fear, because I\u2019m just built different. \nBut to be perfectly honest, spending your early twenties trapped in a foreign country while the FBI tries to put you in jail, sort of just\u2026completely fries your nervous system.\nMy entire fight or flight system is basically now just a single hamster on a wheel.\nOne benefit, though, is its easy to know you\u2019re not making decisions out of fear when you can\u2019t experience any. So, I ended up making the decision to learn new skills and expand my existing subject-matter expertise.\nI still regularly use LLMs, and keep up with new innovations in the space, but I have no intention of pivoting into \u201cAI\u201d right now.\nThe field of doing $stuff with the latest $thing is extremely over-saturated, and I\u2019m quite happy working on the cutting edge of existing technologies. But what I believe to be the biggest harm is not fear, but the downstream effects of the rush to adopt LLMs. The Snake Is Already Eating Its Own Tail LLM Self-Cannibalization One of my other main arguments for why I believe LLMs have peaked, is source cannibalization.\nSince LLMs do not think nor reason, they are heavily reliant on large corpuses of human-produced data for training and RAG. But when LLMs deprive data publishers of revenue either indirectly via training on their works without permission, or directly as a result of straight plagiarism via RAG, this forces publishers behind paywalls.\nThe paywalls not only limits the LLMs\u2019 access to future training data, and it\u2019s ability to use RAG, but also negatively impacts regular humans who are not using LLMs at all. Right now big tech companies operate in a temporary utopia where they\u2019ve been able to capitalize on mass-scale copyright infringement as a result of the free and open internet, but have not yet started to suffer the consequences of the damage they\u2019re causing to the information ecosystem. LLMs act as a sort of magic funnel where users only see the output, not the incalculable amounts of high-quality human-produced data which had to be input.\nAs such, it\u2019s likely people significantly overestimate how much their work (prompting) contributed to the output they received, and grossly underestimate how much of other peoples\u2019 work was required to make it possible.\nIt\u2019s classic egocentric bias. This kind of bias leads to people ignoring the threat LLMs pose to their own data sources.\nThe problem is further amplified by AI slop (low-quality AI generated content), which floods the internet, degrading the average quality of information.\nThis slop not only makes it harder for humans to find high-quality sources, but harder to train LLMs, since allowing the slop to enter LLMs datasets risks creating a feedback loop which could cause the LLM to undergo model collapse. Faux Productivity & LLM Addiction LLMs inherently hijacking the human brains\u2019 reward system.\nBy allowing people to quickly summarize & manipulate the work of others, LLMs use gives the same feeling of achievement one would get from doing the work themselves, but without any of the heavy lifting. The brain is naturally very fragile to instant gratification, which is also part of the mechanism behind drug addiction.\nWhen feelings of accomplishment are tied to completion of a task, the less time taken to accomplish the task, the more frequent the dopamine hits.\nSimulation games often exploit this by reproducing real world tasks, but in a way where they can be completed with much less effort. The most extreme example of short-circuiting the brain\u2019s reward system, is of course drug use. \nBy consuming chemicals which force the brain to release neurotransmitters associated with feelings of accomplishment, users can generate the same feelings of success, without necessarily needing accomplishing any tasks at all. Adderall Studies & AI A while back I encountered several studies researching the effects of Adderall on neurotypicals.\nBoth people with and without ADHD tend to report a significant increase in productivity resulting from taking Adderall.\nIt\u2019s well establish that Adderall boosts productivity in people with ADHD, likely by correcting counteracting their brain\u2019s natural deficit of dopamine and norepinephrine. With neurotypicals on the other hand, the results were very different. \nOne study showed that neurotypicals felt more productive when taking Adderall vs a placebo. But their objective productivity remained unchanged or even declined while under its effects.\nAnother study showed that Adderall use led to a noticeable decline in objective productivity. Since most people without ADHD don\u2019t have deficits in dopamine or norepinephrine, the Adderall increases neurotransmitter levels above normal, producing a high.\nSince dopamine and norepinephrine play a significant role in feelings of confidence, satisfaction, and gratification; \nit\u2019s not really unexpected that a surplus would skew judgement. Having ADHD myself, and having on many occasions accidentally double dosed my Adderall, I\u2019ve personally experienced both sides of this.\nThe genuine productivity boost from using a much-needed medication, and the overstimulated overconfident word soup, which I look back on at a later date with dismay. Whenever I come across yet another fart-huffing self-aggrandizing take on LinkedIn about how programming is dead or LLMs are replacing cows and disrupting big milk, I think about the Adderall studies.\nThese are not the words of a rational person objectively evaluating a new technology, but someone high out of their mind as the result of an LLM-induced dopamine overload.\nIt evokes the exact same feeling of talking to someone who is high on cocaine. Current Research What\u2019s interesting is studies attempting to measure productivity increase due to use of LLMs are actually finding the opposite. \nEveryone feels more productive, but the data is showing a notable decrease in objective productivity among LLM users.\nMy very un-scientific hypothesis is that many LLM users are simply just completely cracked out on dopamine.\nThe euphoria resulting from their perceived now limitless abilities is clouding their judgement. Which makes me wonder: what if we replicated the Adderall study with LLMs?\nWould we find similar results that LLM use does boost productivity for people with ADHD by increasing dopamine and reducing heavy lifting?\nOr is any increase in productivity negated by the fact that low-quality output is inherent to the LLM, and not purely tied the mental state of the user? Either way, the current research perfectly lines up with what I\u2019ve been observing. \nA whole lot of hyperbolic claims from people who just made their first totally not going fail B2B SaaS purely with vibe coding, but not a whole lot of substance.\nI think people are simply overestimating their productivity and abilities as the result of a dopamine high produced by their instant gratification machine. \u201cYou Won\u2019t Be Replaced By AI But By An Employee Using AI\u201d The popular wisdom that\u2019s seen as somewhat of a middle ground between \u201cLLMs are useless plagiarism machines\u201d and \u201cLLMs are going to replace everything ever\u201d is the hypothetical AI-accelerated employee.\nWhile not necessarily terrible advice, I feel like the mainstream interpretation is the opposite of what the advice should be. The fact of the matter is, \u201cprompt engineering\u201d or whatever they\u2019re calling it these days, has a skill cap that\u2019s in the floor.\nPrompting LLMs simply just isn\u2019t a skill, no matter what influencers who definitely weren\u2019t previously claiming Bored Apes are the new Mona Lisa say.\nIf you look at the prompts even the LLM developers themselves are using its things like \u201cplease don\u2019t make stuff up\u201d or \u201cthink extra hard before you answer\u201d. In fact, I\u2019d make a strong argument that what you shouldn\u2019t be doing is \u2018learning\u2019 to do everything with AI.\nWhat you should be doing is learning regular skills.\nBeing a domain expert prompting an LLM badly is going to give you infinitely better results than a layperson with a \u2018World\u2019s Best Prompt Engineer\u2019 mug. The advice is completely backwards, and leads people towards over-reliance on AI, which brings me to the final and most serious issue. LLM Over-Reliance and Cognitive Decline LLMs are somewhat like lossy compression of the entire internet. They boil nuanced topics down into a form that isn\u2019t quite layperson level, but loses a lot of nuance while still evoking a feeling of complete understanding.\nThe part that\u2019s missing is, you don\u2019t know what you don\u2019t know. If you lack an intricate understanding of the task you\u2019re using an LLM for, you have no idea what nuance was lost, or worse, what facts it made up. But I\u2019d actually go a step further and bet $1,000 that in the next 5 years we\u2019ll start to see an abundance of studies showing overuse of LLMs actually results in significant cognitive decline.\nThe brain is much like a muscle in the sense that neural pathways have to be continuously re-enforced through mental exercises.\nI suspect many of us have had the experience of not using a skill long enough to completely un-learn it. LLMs enable this, but with every skill. Maintaining skills isn\u2019t just a case of accessing knowledge regularly, but interacting with it in different ways.\nThere\u2019s a reason why language learning tools have you say words, translate them, read them, write them, and use them in sentences.\nEvery distinct means by which you apply knowledge or a skill further reinforces it and deepens your overall understanding. When people use LLMs, they aren\u2019t just being presented with flimsy surface level understandings of topics.\nThey\u2019re often outsourcing many of their means of reinforcing knowledge to the AI too.\nAnd in some cases, their logic and reasoning itself.\nThe more people lean on LLMs, the more likely they are limit their knowledge expansion, undergo skill regression, and weaken their logic and reasoning ability. So, in fear of being replaced by the hypothetical \u2018AI-accelerated employee\u2019, people are forgoing acquiring essential skills and deep knowledge, instead choosing to focus on \u201cprompt engineering\u201d.\nIt\u2019s somewhat ironic, because if AGI happens there will be no need for \u2018prompt-engineers\u2019. And if it doesn\u2019t, the people with only surface level knowledge who cannot perform tasks without the help of AI will be extremely abundant, and thus extremely replaceable. You can learn how to prompt LLMs at any time, but you can\u2019t learn a decade of specialized skills in an afternoon. Final Thoughts So yes, while I may come off as a massive LLM hater, I feel like I have my reasons.\nWith that said, I am still actively researching and experimenting with LLM regularly, and I\u2019m always open to being proven wrong.\nBut currently, I\u2019m simply not seeing it. I\u2019m not seeing heaps of successful LLM products, businesses, or use cases.\nWhat I\u2019m seeing is a lot of shovel selling, and a huge black hole for VC money. Maybe some day I\u2019ll write a post about the viability of LLMs for something I\u2019m building. \nBut it won\u2019t be today, this year, or likely anytime soon. \nIn fact, I\u2019m currently still getting job offers for manual reverse engineering jobs.\nIt\u2019s extremely common for security companies that use machine learning to hire manual analysts.\nML models need constant tweaking and updating, which means a huge market for experts who can be a part of that process. I\u2019d expect this is where LLMs will go should the tech take off. \nNot job replacement, but a shift towards professionals using their experience to fine tune LLMs, instead of doing the work directly.\nIn fact, I\u2019ve started getting the same offers to consult for LLM companies that I am for traditional ML ones. Times change, but technology progresses slowly. Share this article (https://www.linkedin.com/sharing/share-offsite/?url=https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html&title=Every%20Reason%20Why%20I%20Hate%20AI%20and%20You%20Should%20Too) (Share on LinkedIn) LinkedIn (https://bsky.app/intent/compose?text=https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html&title=Every%20Reason%20Why%20I%20Hate%20AI%20and%20You%20Should%20Too) (Share on LinkedIn) Bluesky Show Comments Please enable JavaScript to view the (https://disqus.com/?ref_noscript) comments powered by Disqus. Stay Informed Subscribe to my newsletter or get notified of new posts. Email address (Email Address) (1) (1) Subscribe (https://marcushutchins.com/) (https://marcushutchins.com/) Marcus Hutchins Threat intelligence analyst, programmer, ex-hacker. (https://www.linkedin.com/in/MalwareTech) (LinkedIn) (https://bsky.app/profile/malwaretech.com) (BlueSky) (https://infosec.exchange/@malwaretech) (Mastodon) (https://youtube.com/MalwareTechBlog) (YouTube) (https://tiktok.com/@malwaretech) (TikTok) (https://instagram.com/malwaretech) (Instagram) (https://twitch.tv/MalwareTechBlog) (Twitch) Featured Posts (/2025/08/every-reason-why-i-hate-ai.html) Aug 4, 2025 (/2025/08/every-reason-why-i-hate-ai.html) Every Reason Why I Hate AI and You Should Too (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) Mar 28, 2025 (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) The US Needs A New Cybersecurity Strategy: More Offensive Cyber Operations Isn't It (/2024/08/exploiting-CVE-2024-38063.html) Aug 27, 2024 (/2024/08/exploiting-CVE-2024-38063.html) CVE-2024-38063 - Remotely Exploiting The Kernel Via IPv6 (/2024/02/bypassing-edrs-with-edr-preload.html) Feb 13, 2024 (/2024/02/bypassing-edrs-with-edr-preload.html) Bypassing EDRs With EDR-Preloading (/2023/12/silly-edr-bypasses-and-where-to-find-them.html) Dec 27, 2023 (/2023/12/silly-edr-bypasses-and-where-to-find-them.html) Silly EDR Bypasses and Where To Find Them (/2023/12/an-introduction-to-bypassing-user-mode-edr-hooks.html) Dec 25, 2023 (/2023/12/an-introduction-to-bypassing-user-mode-edr-hooks.html) An Introduction to Bypassing User Mode EDR Hooks (/2020/12/how-i-found-my-first-ever-zeroday-in-rdp.html) Dec 31, 2020 (/2020/12/how-i-found-my-first-ever-zeroday-in-rdp.html) How I Found My First Ever ZeroDay (In RDP) (/2018/03/best-programming-languages-to-learn-for-malware-analysis.html) Mar 19, 2018 (/2018/03/best-programming-languages-to-learn-for-malware-analysis.html) Best Languages to Learn for Malware Analysis (/2017/05/how-to-accidentally-stop-a-global-cyber-attacks.html) May 13, 2017 (/2017/05/how-to-accidentally-stop-a-global-cyber-attacks.html) How to Accidentally Stop a Global Cyber Attacks (/2015/04/hard-disk-firmware-hacking-part-1.html) Apr 13, 2015 (/2015/04/hard-disk-firmware-hacking-part-1.html) Hard Disk Firmware Hacking (Part 1) Explore Topics (/tag/explainers) Explainers 14 (/tag/malware) Malware 15 (/tag/windows-internals) Windows Internals 10 (/tag/hacking) Hacking 13 (/tag/vulnerability-research) Vulnerability Research 11 (/tag/news) News 10 (/tag/analysis) Analysis 10 (/tag/malware-analysis) Malware Analysis 16 (/tag/programming) Programming 4 (/tag/threat-intelligence) Threat Intelligence 13 (/tag/opinions) Opinions 12 (/tag/stories) Stories 3 (/tag/wannacry) WannaCry 2 (/tag/videos) Videos 3 (/tag/artificial-intelligence) Artificial Intelligence 1 (/tag/technology) Technology 1 You may also like (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) (The US Needs A New Cybersecurity Strategy: More Offensive Cyber Operations Isn't It) Mar 28, 2025 (/tag/opinions) Opinions (/tag/hacking) Hacking (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) The US Needs A New Cybersecurity Strategy: More Offensive Cyber Operations Isn't It (/2023/09/it-might-be-time-to-rethink-phishing-awareness.html) (It might Be Time to Rethink Phishing Awareness) Sep 20, 2023 (/tag/opinions) Opinions (/2023/09/it-might-be-time-to-rethink-phishing-awareness.html) It might Be Time to Rethink Phishing Awareness (/2023/02/a-realistic-look-at-chatgpt-cybercrime.html) (A Realistic Look at Implications of ChatGPT for Cybercrime) Feb 21, 2023 (/tag/malware) Malware (/tag/hacking) Hacking (/tag/opinions) Opinions (/2023/02/a-realistic-look-at-chatgpt-cybercrime.html) A Realistic Look at Implications of ChatGPT for Cybercrime (Top) Menu (/) Home (https://marcushutchins.com/speaking) Speaking (/labs) Labs (https://discord.gg/malwaretech) Discord (https://marcushutchins.com/about) About Me (/contact) Contact Recent Posts (/2025/08/every-reason-why-i-hate-ai.html) Aug 4, 2025 (/2025/08/every-reason-why-i-hate-ai.html) Every Reason Why I Hate AI and You Should Too (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) Mar 28, 2025 (/2025/03/the-us-needs-a-new-cybersecurity-strategy.html) The US Needs A New Cybersecurity Strategy: More Offensive Cyber Operations Isn't It Stay Informed Subscribe to my newsletter or get notified of new posts. Email address (Email Address) Subscribe 2025 \u00a9 (/) MalwareTech (https://www.linkedin.com/in/MalwareTech) (LinkedIn) (https://bsky.app/profile/malwaretech.com) (BlueSky) (https://infosec.exchange/@malwaretech) (Mastodon) (https://youtube.com/MalwareTechBlog) (YouTube) (https://tiktok.com/@malwaretech) (TikTok) (https://instagram.com/malwaretech) (Instagram) (https://twitch.tv/MalwareTechBlog) (Twitch) " ], "output": "htmltotext.txt", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:42.087239+00:00", "status": "succeeded" } ], "media": [ { "cmd": [ "/usr/local/bin/yt-dlp", "--restrict-filenames", "--trim-filenames", "128", "--write-description", "--write-info-json", "--write-annotations", "--write-thumbnail", "--no-call-home", "--write-sub", "--write-auto-subs", "--convert-subs=srt", "--yes-playlist", "--continue", "--no-abort-on-error", "--ignore-errors", "--geo-bypass", "--add-metadata", "--format=(bv*+ba/b)[filesize<=750m][filesize_approx<=?750m]/(bv*+ba/b)", "--skip-download", "--cache-dir=/data/yt-dlp-cache/", "--cookies=/data/yt-dlp-cache/cookies.txt", "--proxy=socks5://tor-socks-proxy:9150", "--no-playlist", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "2024.10.7", "end_ts": "2025-08-10T01:52:49.788318+00:00", "index_texts": [], "output": "media/", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:44.264861+00:00", "status": "succeeded" } ], "mercury": [ { "cmd": [ "/home/archivebox/.npm/bin/postlight-parser", "--format=text", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "2.2.3", "end_ts": "2025-08-10T01:52:42.040316+00:00", "index_texts": null, "output": "ArchiveError: Mercury was not able to get article text from the URL", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:41.191435+00:00", "status": "failed" } ], "pdf": [], "readability": [ { "cmd": [ "/home/archivebox/.npm/bin/readability-extractor", "/tmp/tmp8k1f5siy", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "0.0.11", "end_ts": "2025-08-10T01:52:32.393273+00:00", "index_texts": [ "One accusation that tends to get thrown my way often is being anti-innovation.\nIf you\u2019ve followed me for any amount of time, you\u2019ve probably seen me dunking on everything from Cryptocurrency to Large Language Models. \nThis often leads to people asking the question \u201cwhy is someone so technical so anti-technology?\u201d.\n\nTo a lot of people on social media, I come of as just some unhinged weirdo who posts their every half-baked thought out onto the internet. This is indeed correct. \nBut when it comes to my research, work, and career, I\u2019m extremely calculated and deliberate in the decisions I make.\n\nThe reason I\u2019m not diving head first into everything AI isn\u2019t because I fear it or don\u2019t understand it, it\u2019s because I\u2019ve already long since come to my conclusion about the technology.\nI\u2019m neither of the opinion that it\u2019s completely useless or revolutionary, simply that the game being played is one I neither currently need nor want to be a part of.\n\nAI Mania\nOne thing that\u2019s certain is that Generative AI is in a bubble. \nThat\u2019s not to say AI as a technology will pop, or that there isn\u2019t genuine room for a lot more growth; simply, the level of hype far outweighs the current value of the tech.\n\nMost (reasonable) people I speak to are of one of three opinions:\n\n These technologies are fundamentally unsustainable and the hype will be short-lived.\n There will be some future breakthrough that will bring the technology in line with the hype, but in the meantime everyone is essentially just relying on creative marketing to keep the money flowing.\n The tech has a narrow use case for which they are exceedingly valuable, but almost everything else is just hype.\n\n\nWhenever I\u2019m critical of anything GenAI, without fail I get asked the same question.\n\u201cdo you think every major CEO could be wrong?\u201d\n\nThe answer to that is: yes. History is littered with examples of industry titans going nuts, losing more money than the GDP of an entire country, saying \u201clol, my bad\u201d, then finding something else to do.\n\nI grew up during the fallout from the great financial crisis. I watched first hand as the biggest most prestigious financial institution crashed the entire global economy.\nTurns out, in the short term playing hot potato with debt derivatives backed by imaginary money and fraud is a great business model. In the long term, not so much.\n\nIt\u2019s not even necessarily that corporate executives are being stupid.\nSometimes they are, which can result in things like sinking more money that it cost the US government to put the sun in a bomb into the worst VR game ever. But usually it\u2019s just greed and shortsightedness.\n\nAI right now feels much the same. An industry fueled by the gluttony of myopic visionaries. An industry grasping at every straw to find a use case for their technology. An industry built on the premise of hyperbole and empty promises. \nBut in this specific case, I actually don\u2019t think big tech companies are making the wrong decision, at least, considering the choices they have available.\n\nThe Big Tech AI Hedge Bet\nThe biggest threat to tech companies right now is AGI (Artificial General Intelligence). \nAGI is a theoretical AI model which surpasses humans on their ability to learn, reason, think, and adapt.\nThe risk is that if one company were to figure out AGI, they\u2019d have an extreme competitive advantage over then rest in almost every space.\n\nSinking 100 billion dollars into AI research isn\u2019t going to kill any big tech company.\nIf things don\u2019t pan out, it won\u2019t even matter. Every other company will have lost similar amounts of money chasing ghosts.\nBut if AGI does happen, the market will shift so fast and so significantly that any major player without AGI will be left in the dust.\nQuite simply, the possibility of AGI is an existential threat to big tech.\n\nSo from an executive perspective, lighting comically large piles of money on fire trying to teach graphics cards how to read is, surprisingly, the logical play.\nThe rest, well, that\u2019s all just creative marketing. It\u2019s very difficult to show up to a quarterly shareholder meeting and tell your investors you just vaporized another $10 billion for absolutely no return-on-investment. At least, that is, without them questioning if you\u2019ve completely lost your mind.\nWhich is where leaning into the hype plays into it.\n\nTesla was able to keep investment flowing for over a decade by claiming each year that full self-driving was coming the very next year.\nBig Tech can easily do the same. \n\u201c90% of our code is now AI.\u201d, \u201cwe\u2019re seeing spark of AGI.\u201d, \u201cwe\u2019re so worried that our super smart word processor will become sentient and kill everyone\u201d.\nIt\u2019s all just vapid bluster designed to keep investors on board long enough for them to hopefully figure out a path to AGI.\nFake it till you make it, but at an entire industry-wide scale.\n\nNow, obviously, without sitting down individually with every big tech CEO, I can\u2019t tell you how many actually genuinely believe they are close to AGI.\nExternally, it\u2019s difficult to tell true innovation from grifting & hubris, especially in the tech space. \nBut based on the outright insanity of many claim being made by major tech CEOs, I\u2019d guess it\u2019s mostly the latter.\n\nWhich brings me on to the downstream effects.\n\nTrickle Down Hypenomics\nWhen you have all these big tech companies releasing audacious claims about AI on a near daily basis, you start getting asked the question \u201cwell, what are WE doing to prepare for AI?\u201d\nThe correct answer is, of course, nothing.\nUnless you have billions upon billions of dollars to build and train your own AI model, you\u2019re basically just a future customer of some big tech company\u2019s expensive subscription service.\n\nThe hugely prohibitive cost to both researching and building Large Language Models makes it the perfect walled garden for any big tech company.\nShould they succeed at AGI, they can sell the subscriptions for slightly less than the equivalent human labor, cornering much of the market.\nShould they fail, they can just look for some other way to lock everyone into yet another overpriced subscription model.\n\nBut in the meantime, why not slap \u201cpowered by AI\u201d on your toilet brush and win over a few extra customers from your competitors.\nThis is essentially what you\u2019re seeing now with every single company just ham fisting some half-baked LLM chatbot into their workflow.\n\nThen, of course, there\u2019s the outright grifters. \nThe people who jump on the latest hype train, try to position themselves as visionaries in that space, then take their bag of cash and leave right before it all comes crashing down.\nIt\u2019s no surprise that half of the AI influencer were previously hawking NFTs, or trying to sell you blockchain toasters.\n\nThe Apple Approach\nWhat I find most interesting though, is Apple. I\u2019ve always been a fan of Apple, just not in the typical sense.\nI don\u2019t care much for iPhones or MacBooks, but I\u2019ve always been fascinated by the way the company operates.\nThey tend to steer clear of grasping at straws to find new business models, massively over-hiring and over-firing, or immediately hopping on every new trend.\nThey tend to sit back, calculate, then make very careful and deliberate moves.\n\nApple initially began exploring the LLM space, performing internal research, and was even in talks to invest in OpenAI.\nThey later released two studies on LLM reasoning (one in October 2024, and a second in June 2025). Both studies argued that LLMs do not reason, they simply perform statistical pattern matching, which they pass off as reasoning.\nWithout true reasoning ability, LLMs will never become AGI, which appears to be Apple\u2019s current public stance.\n\nWhile Apple is still a major player in the AI space, they typically lean towards a preference for traditional AI.\nThey\u2019re almost certainly still doing LLM research in the background, but they\u2019ve decided against dropping everything to go full throttle on the hype train.\nIronically, this has started to make their investors, who have bought into the hype, quite upset.\nPeople are starting to question \u201cwhy isn\u2019t Apple doing things with AI\u201d.\n\nIt may very well be the case that Apple too finds themselves pressured into going all out on LLM mania. \nDespite being both calculated and cautious enough to avoid over-committing to AI, while still mitigating any risks posed by potential future AGI, Apple is still subject to the whims of its investors, who really love buzzwords.\n\nOf course, language translation is not AI, Siri is not AI, image identification & classification is not AI, the only acceptable form of AI is shoving some half-baked LLM chatbot into somewhere it doesn\u2019t belong.\nAfter all, how can we possibly be sure that Apple is \u201cdoing AI\u201d if they aren\u2019t boiling the ocean by shoving something like Copilot into every god damn app.\n\nLLMs Are Not The Path To AGI\nApple\u2019s conclusion is the same conclusion I came to very early on in my experimentation with LLMs.\nI don\u2019t say this as some attempt to flex \u201cI was here first\u201d, because I\u2019m not a 3 trillion dollar tech company with a lot to lose.\nThe more important point is this whole time I have been operating from the position that LLMs are not, and never will be AGI.\nMy position has remained completely unchanged.\nAs such, if it sounds like I might\u2019ve received psychic damage from having to listen to 5 years of non-stop AI drivel on LinkedIn, it\u2019s because that is very much the case.\n\nReasoning has long been a very contentious topic among LLM proponents.\nThis is heavily fueled by the fact these models are somewhat of a black box. \nWe can\u2019t just take a peek inside and draw conclusions about how or why it produced the output it did.\nEven the people who build LLMs for a living readily admit this.\n\nLLMs (and Neural Networks in general) are very similar to the human brain in that we can explain what individual neurons do, but put enough of them together and nobody has any clue what caused it to do the thing it did.\nEven the best neuroscientists have very little understand of how the brain works as a whole.\n\nThis is where things like the \u2018Stochastic Parrot\u2019 or \u2018Chinese room\u2019 arguments comes in.\nTrue reasoning is only one of many theories as to how LLMs produce the output they do; it\u2019s also the one which requires the most assumptions (see: Occam\u2019s Razor).\nAll current LLM capabilities can be explained by much more simplistic phenomena, which fall far short of thinking or reasoning.\n\nIf a machine is consuming and transforming incalculable amounts of training data produced by humans, discussed by humans, and explained by humans.\nWhy would the output not look identical to human reasoning?\nIf I were to photocopy this article, nobody would argue that my photocopier wrote it and therefore can think.\nBut add enough convolutedness to the process, and it looks a lot like maybe it did and can.\n\nTrue Reasoning vs Statistical Pattern Matching\nFor me, one of the clear distinctions between true reasoning and pattern matching, is what happens when you remove access to new information.\nMany argue that LLMs are not plagiarism machines, they learn like humans do. \nThey consume knowledge from teachers and books, developing an understanding along the way.\n\nAs a professional researcher, I too learned most of what I know by reading the works of far greater researchers.\nBut there became a point where I knew enough to perform my own original research in uncharted waters.\nI now can and regularly do research topics where there is no other source to check my work against.\nThis is not something LLMs are good at, or debatably, can even do at all.\n\nGiving an LLM access to the entire internet and all of recorded human knowledge, then testing them with a quiz designed for humans is obviously just a cheap parlor trick.\nI, too, could score 100% on a multiple-choice exam if you let me Google all the answers.\nThe true measure of LLM intelligence and reasoning should not be refactoring existing information, but the ability to produce truly novel works.\n\nSure, an LLM could probably create a brand-new pop song, because it has plenty of existing songs to analyze, allowing it to produce something seemingly new, but really just based on existing patterns.\nYet, every time I tried to get LLMs to perform novel research, they fail because they don\u2019t have access to existing literature on the topic.\nWhereas, humans, on the other hand, discovered everything humanity knows.\n\nWhere people get tied up is with the argument, \u201cwell most humans don\u2019t produce novel work either\u201d. \nBut this is not because they\u2019re fundamentally incapable of it.\nThe average person is simply just sandbagged with an unfulfilling job.\nIdeally, without the perverse incentives of shareholder value, LLMs would automate all the busy work, allowing humans to focus on more meaningful pursuits.\n\nI\u2019ve made this argument many times before. But if humans could come up with all the groundbreaking discoveries they have, reading only as many books or research papers as they realistically could.\nWhere are all the major LLM discoveries? \nAn individual human may be limited in their ability to make novel discoveries as a result of competing with 8 billion other humans.\nBut LLMs have access to the knowledge of all.\n\nYou\u2019d think that a machine with access to every book ever written, every paper ever published, every speech ever recorded, and every study ever produced could do a lot better than some person who can read maybe 1 book per day.\nYet, nothing. There\u2019s the odd \u201cmaybe the LLM did something novel, but we don\u2019t know yet\u201d posts here and there, but if they could actually think, with as much information as they have, there\u2019d be groundbreaking discoveries literally falling from the sky.\n\nIn reality, all we\u2019ve created is a bot which is almost perfect at mimicking human-like natural language use, and the rest is people just projecting other human qualities on to it.\nQuite simply, \u201cLLMs are doing reasoning\u201d is the \u201clook, my dog is smiling\u201d of technology.\nIn exactly the same way that dogs don\u2019t convey their emotions via human-like facial expressions, there\u2019s no reason to believe that even if computer could think, it\u2019d perfectly mirror what looks like human reasoning.\n\nThe Difficulty Of Disproving LLM Reasoning\nOne of the main challenges with testing LLM reasoning is it usually relies on giving it a novel problem. \nBut as soon as new problems are published, answers are published too, at which point the LLM can just regurgitate an existing answer from its training data.\n\nA logical problem I previously used to tests early LLMs was one called \u201cThe Wolf, The Goat, And The Cabbage\u201d. The problem is simple.\nYou\u2019re walking with a wolf, a goat, and a cabbage. You come to a river which you need to cross.\nThere is a small boat which only has enough space for you and one other item. \nIf left unattended, the wolf will eat the goat, and the goat will eat the cabbage. How do you get all 3 safely across?\n\nThe correct answer is you take the goat across, leaving behind the wolf and the cabbage. \nYou then return and fetch the cabbage, leaving the goat alone on the other side.\nBecause the goat and cabbage cannot be left alone together, you take the goat back, leaving just the cabbage.\nNow, you can take the wolf across, leaving the wolf and the cabbage alone on the other side, finally returning to fetch the goat.\n\nAny LLM could effortlessly answer this problem, because it has thousands of instances of the problem and the correct solution in its training data.\nBut it was found that by simply swapping out one item but keeping the same constraints, the LLM would no longer be able to answer.\nReplacing the wolf with a lion, would result in the LLM going off the rails and just spewing a bunch of nonsense.\n\nThis made it clear the LLM was not actually thinking or reasoning through the problem, simply just regurgitating answers and explanations from its training data.\nAny human, knowing the answer to the original problem, could easily handle the wolf being swapped for a lion, or the cabbage for a lettuce.\nBut LLMs, lacking reasoning, treated this as an entirely new problem.\n\nOver time this issue was fixed. It could be that the LLM developers wrote algorithms to identify variants of the problem.\nIt\u2019s also possible that people posting different variants of the problem allowed the LLM to detect the core pattern, which all variants follow, allowing it to substitute words where needed.\n\nThis is when someone found you could just break the problem, and the LLM\u2019s pattern matching along with it. Either by making it so none of the objects could be left unattended, or all of them could.\nIn some variants there was no reason to cross the river, the boat doesn\u2019t fit anyone, was actually a car, or has enough space to carry all the items at once.\nHumans, having actual logic and reasoning abilities could easily identify the broken versions of the problems and answer accordingly, but the LLMs would just output incoherent gibberish.\n\nBut of course, as more and more ways to disprove LLM reasoning were found, the developers just found ways to fix them. \nI strongly suspect these issues are not being fixed by any introduction of actual logic or reasoning, but by sub-models built to address specific problems.\nIf this is the case, I\u2019d argue we\u2019re moving away from AGI and back towards building problem specific ML models, which is how \u201cAI\u201d has worked for decades.\n\nThe Limitations of LLMs\nI\u2019m personally leaning towards the opinion that LLMs as a technology will soon, or have already, capped out.\nThey fast hit a ceiling where giving them more data, more parameters, and more token stopped leading to any noticeable improvement.\n\nMore recent technological developments, just feel more like hacks.\n\nChain-of-Thought Reasoning (CoT)\nCoT essentially just has the LLM break the problem down into smaller parts.\nTo give an oversimplified answer, if I asked an LLM 1 + 1 + 1. It could just answer 3, based on its training data.\nThough it could also break the problem down into 1 + 1 to get 2. Then it can add 2 + 1 to get 3. \nFor validity, it could then just answer the problem normally, or via different approaches, comparing the results.\n\nThis addresses hallucinations to a certain degree (which are an inherent feature of LLMs, not a bug that can be fixed).\nIt also in some cases enabled LLMs to solve problems that they can\u2019t simply one-shot based on their training data.\nBut it\u2019s still reliant on the LLM being able to break down the problem in the first place, not hallucinate any of the stages, and also takes a ton more time & compute.\n\nSince the appeal of LLMs, for most users, is getting a semi-decent answer as quickly as possible, few people want to wait the several minutes it takes for the LLM to boil the ocean turning 1 prompt into 50.\nIt\u2019s also not really an improvement upon the LLM technology, it\u2019s just solving the fundamental flaws of LLMs by adding more LLMs.\n\nRetrieval-Augmented Generation (RAG)\nAs a researcher and writer, this is by far my least favorite LLM feature.\n\nRAG was designed to at least partially address the issue of the extreme time and computational cost of training LLMs.\nLLMs aren\u2019t re-trained super frequently, which means the datasets quickly becomes stale.\nYou\u2019ll probably recall how ChatGPT used to respond with \u201cas of my knowledge cutoff of , \" before giving the completely wrong answer.\n\nRAG basically allows the LLM to search the internet for fresh data relevant to the user\u2019s query, enabling it to fetch the most up-to-date information.\nThe LLM can then use its training data to summarize the information retrieved via RAG. \nEssentially, this combines LLMs and search engines into a single product.\n\nSo, why do I hate this? Well, it\u2019s basically glorified plagiarism. \nWhile I\u2019d argue LLMs in general are just Plagiarism-as-a-Service, RAG is a lot closer to actual plagiarism that typical LLM behavior.\nAs I\u2019ve already argued, LLMs don\u2019t think or reason.\nThus, all RAG is really doing is using the LLMs\u2019 natural language abilities to summarize or re-word some news article, blog post, or research paper.\nThis deprives the original author of revenue & website traffic, while not transforming their work in any meaningful way.\n\nSince RAG is still just a wrapper that sits on top of the core LLM technology, it\u2019s still vulnerable to hallucinations.\nThe technology also struggles when there\u2019s too few sources, or when \u201cunderstanding\u201d the information would require additional context which doesn\u2019t exist in the LLMs training data.\nThe other major flaw, is the LLM not knowing when to use RAG, or failing to identify stale information, since not all search results will have publish dates.\n\nI ran into this issue very recently during a joke research project in which I gave an LLM several thousand dollars, a brokerage account with option trading enabled, and complete autonomy to place trades.\nI\u2019ll publish the full breakdown once the project has run its course, but one problem it ran into a lot was the LLM not using RAG to fetch current data on stocks.\nThe LLM would just quote stock prices based on whatever they were at the time it was last trained, which in my case was over a year ago.\nThis lead to the bot making trades based upon completely inaccurate price information.\n\nFear Of Replacement & Impulse Decision-Making\nThe Current State Of Tech\nOne thing that has become very clear to me, is that similar to Blockchain, BigData, Cloud, and NFTs, a lot of the activity in the LLM space is motivated by fear.\n\nThe fear itself, is extremely justified. I\u2019ve not seen a job market this brutal in my entire career.\nPreviously, layoffs in the double-digit percentage were something typically reserved for major economic crises or bankruptcies.\nNow, it\u2019s just something massively profitable tech companies do on a whim for seemingly no reason.\n\nI\u2019m regularly getting desperate DMs asking for help finding a job. I\u2019ve had extremely technically talented friends forced to work non-tech jobs to ride out the current market.\nThe recruiters and hiring managers I talk to regularly tell me about having to pull down all their job postings after less than a day because they\u2019ve already gotten thousands of applications.\n\nMuch of this fear is exacerbated by people attributing the mass layoffs and lack of available jobs to AI replacement.\nThis is very much not the case, but the economics of it all is extremely complex and would require its own article.\nBut what matter is that people believe it to be true, and tech companies are more than happy to lean into those narratives to hype up their AI products.\n\nBut fear makes people behave irrationally, and that\u2019s currently a main driving force behind AI adoption.\n\nThe LLM Train Is Not Leaving The Station\nProbably one of the most common fallacies I see in tech is that it pays to be early. The \u201cfirst mover advantage\u201d.\nThis is not something that has ever seemed consistent with reality.\nWhen a new technology comes around, it\u2019s riddled with flaws that need to be ironed out.\nThe use cases aren\u2019t clear, the viability of the technology isn\u2019t clear, It\u2019s a whole lot of trial-and-error.\n\nBut people still believe there is a benefit to being early. \nI suspect it\u2019s due to the fact you only hear about the few companies who adopted a new technology early and became successful, not the thousands that failed along the way.\nI\u2019d argue that far more companies succeed being late than early.\nThey can analyze where others went wrong. They can look for gaps in the market. They build on what\u2019s already been done.\nRunning blindfolded into a minefield is simply not a good business model in my books. But many seem to think it is, at least when you call it \u201cAI\u201d.\n\nA Bit Of History\nThe first web search engine, \u201cArchie\u201d was launched in 1990. Yahoo launched in 1995 along with AltaVista. Dogpile and AskJeeves came about in 1996, and AOL in 1997.\nGoogle, the current industry titan, only entered the race in 1998. They weren\u2019t first, they weren\u2019t early, they just did it better.\n\nThe same is true for Apple. When they announced the iPhone, most of the incumbents balked at it.\nThey were certain it wasn\u2019t a viable product. It went against the current conventional wisdom. \nIn 2022, Apple became the first company on earth to reach a 3 trillion dollar valuation.\n\nTesla was over a century late to the electric car game, building a product long concluded to be of no interest to consumers.\nIt\u2019s now worth more than basically every car company combined. Objectively it\u2019s not actually worth that, most of the investors are just drunk, but Tesla did successfully create a market for EVs.\n\nI could go on for days and days. But simply put, first mover advantage is for board games and patent applications, not adopting new technologies.\n\nWhy I Have No Interest In Showing Up to The Party Early\nRight now, LLMs are an extremely immature technology. \nI personally believe they\u2019re not going to get much better than this, but a breakthrough innovation could change that.\nEither way, it doesn\u2019t matter to me. If the technology is a fad and completely implodes, I couldn\u2019t care less.\nIf it\u2019s not a bubble and LLMs actually turn out to be the new best thing, I can easily adopt them into my own business model.\n\nI\u2019m professionally late to every party. I learned Assembly language in 2008. \nMalware reversing engineering in 2011. \nVulnerability Research in 2014.\nThis blog, mostly documenting manual malware analysis, something that has been ML automated since before I was born, is what made my career.\n\nSo when I see people jumping on the latest hype, telling me I\u2019m going to get left behind, I can only chuckle.\nIf LLMs as a technology are viable, they\u2019ll still be around when and if I decide they\u2019re useful for me.\nIf not, I\u2019ll have missed out on losing my life\u2019s savings in Beanie Babies, The DotCom bubble, or NFTs.\n\nPart of me wishes I could just claim that I\u2019m not motivated by fear, because I\u2019m just built different. \nBut to be perfectly honest, spending your early twenties trapped in a foreign country while the FBI tries to put you in jail, sort of just\u2026completely fries your nervous system.\nMy entire fight or flight system is basically now just a single hamster on a wheel.\nOne benefit, though, is its easy to know you\u2019re not making decisions out of fear when you can\u2019t experience any.\n\nSo, I ended up making the decision to learn new skills and expand my existing subject-matter expertise.\nI still regularly use LLMs, and keep up with new innovations in the space, but I have no intention of pivoting into \u201cAI\u201d right now.\nThe field of doing $stuff with the latest $thing is extremely over-saturated, and I\u2019m quite happy working on the cutting edge of existing technologies.\n\nBut what I believe to be the biggest harm is not fear, but the downstream effects of the rush to adopt LLMs.\n\nThe Snake Is Already Eating Its Own Tail\nLLM Self-Cannibalization\nOne of my other main arguments for why I believe LLMs have peaked, is source cannibalization.\nSince LLMs do not think nor reason, they are heavily reliant on large corpuses of human-produced data for training and RAG.\n\nBut when LLMs deprive data publishers of revenue either indirectly via training on their works without permission, or directly as a result of straight plagiarism via RAG, this forces publishers behind paywalls.\nThe paywalls not only limits the LLMs\u2019 access to future training data, and it\u2019s ability to use RAG, but also negatively impacts regular humans who are not using LLMs at all.\n\nRight now big tech companies operate in a temporary utopia where they\u2019ve been able to capitalize on mass-scale copyright infringement as a result of the free and open internet, but have not yet started to suffer the consequences of the damage they\u2019re causing to the information ecosystem.\n\nLLMs act as a sort of magic funnel where users only see the output, not the incalculable amounts of high-quality human-produced data which had to be input.\nAs such, it\u2019s likely people significantly overestimate how much their work (prompting) contributed to the output they received, and grossly underestimate how much of other peoples\u2019 work was required to make it possible.\nIt\u2019s classic egocentric bias.\n\nThis kind of bias leads to people ignoring the threat LLMs pose to their own data sources.\nThe problem is further amplified by AI slop (low-quality AI generated content), which floods the internet, degrading the average quality of information.\nThis slop not only makes it harder for humans to find high-quality sources, but harder to train LLMs, since allowing the slop to enter LLMs datasets risks creating a feedback loop which could cause the LLM to undergo model collapse.\n\nFaux Productivity & LLM Addiction\nLLMs inherently hijacking the human brains\u2019 reward system.\nBy allowing people to quickly summarize & manipulate the work of others, LLMs use gives the same feeling of achievement one would get from doing the work themselves, but without any of the heavy lifting.\n\nThe brain is naturally very fragile to instant gratification, which is also part of the mechanism behind drug addiction.\nWhen feelings of accomplishment are tied to completion of a task, the less time taken to accomplish the task, the more frequent the dopamine hits.\nSimulation games often exploit this by reproducing real world tasks, but in a way where they can be completed with much less effort.\n\nThe most extreme example of short-circuiting the brain\u2019s reward system, is of course drug use. \nBy consuming chemicals which force the brain to release neurotransmitters associated with feelings of accomplishment, users can generate the same feelings of success, without necessarily needing accomplishing any tasks at all.\n\nAdderall Studies & AI\nA while back I encountered several studies researching the effects of Adderall on neurotypicals.\nBoth people with and without ADHD tend to report a significant increase in productivity resulting from taking Adderall.\nIt\u2019s well establish that Adderall boosts productivity in people with ADHD, likely by correcting counteracting their brain\u2019s natural deficit of dopamine and norepinephrine.\n\nWith neurotypicals on the other hand, the results were very different. \nOne study showed that neurotypicals felt more productive when taking Adderall vs a placebo. But their objective productivity remained unchanged or even declined while under its effects.\nAnother study showed that Adderall use led to a noticeable decline in objective productivity.\n\nSince most people without ADHD don\u2019t have deficits in dopamine or norepinephrine, the Adderall increases neurotransmitter levels above normal, producing a high.\nSince dopamine and norepinephrine play a significant role in feelings of confidence, satisfaction, and gratification; \nit\u2019s not really unexpected that a surplus would skew judgement.\n\nHaving ADHD myself, and having on many occasions accidentally double dosed my Adderall, I\u2019ve personally experienced both sides of this.\nThe genuine productivity boost from using a much-needed medication, and the overstimulated overconfident word soup, which I look back on at a later date with dismay.\n\nWhenever I come across yet another fart-huffing self-aggrandizing take on LinkedIn about how programming is dead or LLMs are replacing cows and disrupting big milk, I think about the Adderall studies.\nThese are not the words of a rational person objectively evaluating a new technology, but someone high out of their mind as the result of an LLM-induced dopamine overload.\nIt evokes the exact same feeling of talking to someone who is high on cocaine.\n\nCurrent Research\nWhat\u2019s interesting is studies attempting to measure productivity increase due to use of LLMs are actually finding the opposite. \nEveryone feels more productive, but the data is showing a notable decrease in objective productivity among LLM users.\nMy very un-scientific hypothesis is that many LLM users are simply just completely cracked out on dopamine.\nThe euphoria resulting from their perceived now limitless abilities is clouding their judgement.\n\nWhich makes me wonder: what if we replicated the Adderall study with LLMs?\nWould we find similar results that LLM use does boost productivity for people with ADHD by increasing dopamine and reducing heavy lifting?\nOr is any increase in productivity negated by the fact that low-quality output is inherent to the LLM, and not purely tied the mental state of the user?\n\nEither way, the current research perfectly lines up with what I\u2019ve been observing. \nA whole lot of hyperbolic claims from people who just made their first totally not going fail B2B SaaS purely with vibe coding, but not a whole lot of substance.\nI think people are simply overestimating their productivity and abilities as the result of a dopamine high produced by their instant gratification machine.\n\n\u201cYou Won\u2019t Be Replaced By AI But By An Employee Using AI\u201d\nThe popular wisdom that\u2019s seen as somewhat of a middle ground between \u201cLLMs are useless plagiarism machines\u201d and \u201cLLMs are going to replace everything ever\u201d is the hypothetical AI-accelerated employee.\nWhile not necessarily terrible advice, I feel like the mainstream interpretation is the opposite of what the advice should be.\n\nThe fact of the matter is, \u201cprompt engineering\u201d or whatever they\u2019re calling it these days, has a skill cap that\u2019s in the floor.\nPrompting LLMs simply just isn\u2019t a skill, no matter what influencers who definitely weren\u2019t previously claiming Bored Apes are the new Mona Lisa say.\nIf you look at the prompts even the LLM developers themselves are using its things like \u201cplease don\u2019t make stuff up\u201d or \u201cthink extra hard before you answer\u201d.\n\nIn fact, I\u2019d make a strong argument that what you shouldn\u2019t be doing is \u2018learning\u2019 to do everything with AI.\nWhat you should be doing is learning regular skills.\nBeing a domain expert prompting an LLM badly is going to give you infinitely better results than a layperson with a \u2018World\u2019s Best Prompt Engineer\u2019 mug.\n\nThe advice is completely backwards, and leads people towards over-reliance on AI, which brings me to the final and most serious issue.\n\nLLM Over-Reliance and Cognitive Decline\nLLMs are somewhat like lossy compression of the entire internet. They boil nuanced topics down into a form that isn\u2019t quite layperson level, but loses a lot of nuance while still evoking a feeling of complete understanding.\nThe part that\u2019s missing is, you don\u2019t know what you don\u2019t know. If you lack an intricate understanding of the task you\u2019re using an LLM for, you have no idea what nuance was lost, or worse, what facts it made up.\n\nBut I\u2019d actually go a step further and bet $1,000 that in the next 5 years we\u2019ll start to see an abundance of studies showing overuse of LLMs actually results in significant cognitive decline.\nThe brain is much like a muscle in the sense that neural pathways have to be continuously re-enforced through mental exercises.\nI suspect many of us have had the experience of not using a skill long enough to completely un-learn it. LLMs enable this, but with every skill.\n\nMaintaining skills isn\u2019t just a case of accessing knowledge regularly, but interacting with it in different ways.\nThere\u2019s a reason why language learning tools have you say words, translate them, read them, write them, and use them in sentences.\nEvery distinct means by which you apply knowledge or a skill further reinforces it and deepens your overall understanding.\n\nWhen people use LLMs, they aren\u2019t just being presented with flimsy surface level understandings of topics.\nThey\u2019re often outsourcing many of their means of reinforcing knowledge to the AI too.\nAnd in some cases, their logic and reasoning itself.\nThe more people lean on LLMs, the more likely they are limit their knowledge expansion, undergo skill regression, and weaken their logic and reasoning ability.\n\nSo, in fear of being replaced by the hypothetical \u2018AI-accelerated employee\u2019, people are forgoing acquiring essential skills and deep knowledge, instead choosing to focus on \u201cprompt engineering\u201d.\nIt\u2019s somewhat ironic, because if AGI happens there will be no need for \u2018prompt-engineers\u2019. And if it doesn\u2019t, the people with only surface level knowledge who cannot perform tasks without the help of AI will be extremely abundant, and thus extremely replaceable.\n\nYou can learn how to prompt LLMs at any time, but you can\u2019t learn a decade of specialized skills in an afternoon.\n\nFinal Thoughts\nSo yes, while I may come off as a massive LLM hater, I feel like I have my reasons.\nWith that said, I am still actively researching and experimenting with LLM regularly, and I\u2019m always open to being proven wrong.\nBut currently, I\u2019m simply not seeing it. I\u2019m not seeing heaps of successful LLM products, businesses, or use cases.\nWhat I\u2019m seeing is a lot of shovel selling, and a huge black hole for VC money.\n\nMaybe some day I\u2019ll write a post about the viability of LLMs for something I\u2019m building. \nBut it won\u2019t be today, this year, or likely anytime soon. \nIn fact, I\u2019m currently still getting job offers for manual reverse engineering jobs.\nIt\u2019s extremely common for security companies that use machine learning to hire manual analysts.\nML models need constant tweaking and updating, which means a huge market for experts who can be a part of that process.\n\nI\u2019d expect this is where LLMs will go should the tech take off. \nNot job replacement, but a shift towards professionals using their experience to fine tune LLMs, instead of doing the work directly.\nIn fact, I\u2019ve started getting the same offers to consult for LLM companies that I am for traditional ML ones.\n\nTimes change, but technology progresses slowly." ], "output": "readability/", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:26.956365+00:00", "status": "succeeded" } ], "screenshot": [], "singlefile": [], "title": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" ], "cmd_version": "8.10.1", "end_ts": "2025-08-10T01:52:25.113526+00:00", "index_texts": null, "output": "Every Reason Why I Hate AI and You Should Too", "pwd": "/data/archive/1754790714.903636", "schema": "ArchiveResult", "start_ts": "2025-08-10T01:52:24.996230+00:00", "status": "succeeded" } ], "wget": [] }, "icons": null, "is_archived": true, "is_static": false, "latest": { "archive_org": "TimeoutExpired: Command '['/usr/bin/curl', '--silent', '--location', '--compressed', '--proxy', 'socks5://tor-socks-proxy:9150', '--head', '--max-time', '60', '--user-agent', 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)', 'https://web.archive.org/save/https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html']' timed out after 60 seconds", "dom": "output.html", "favicon": "favicon.ico", "git": null, "media": "media/", "pdf": null, "screenshot": null, "singlefile": null, "title": "Every Reason Why I Hate AI and You Should Too", "warc": null, "wget": null }, "link_dir": "/data/archive/1754790714.903636", "newest_archive_date": "2025-08-10T01:52:49.877029+00:00", "num_failures": 2, "num_outputs": 7, "oldest_archive_date": "2025-08-10T01:51:59.851491+00:00", "path": "/2025/08/every-reason-why-i-hate-ai.html", "schema": "Link", "scheme": "https", "snapshot_abid": "snp_01K28TTQJEE53AB9EB01352N38", "snapshot_id": "8501c695-2272-4f25-be2b-2c8806515468", "sources": [ "/data/sources/1754790713-import.txt" ], "tags": null, "tags_str": "", "timestamp": "1754790714.903636", "title": "Every Reason Why I Hate AI and You Should Too", "url": "https://malwaretech.com/2025/08/every-reason-why-i-hate-ai.html" }