47 Comments

1. 

   @slidebean

   March 8, 2025 at 11:40 pm

   So, guess we’ll still have to count the R’s in ‘strawberry’ ourselves… 🍓
   What else do you think AI still can’t do well (yet)?

   Also, don’t forget—Use code slidebean for a free 3-month trial of NordPass Business. No credit card required! 👉 https://nordpass.com/slidebean

   Reply
2. 

   @KingLouisXIV

   March 8, 2025 at 11:40 pm

   sorry. the typing sounds made me dip out

   Reply
3. 

   @LocomotiveThought

   March 8, 2025 at 11:40 pm

   This is like the cassette to the record, the DVD to the VHS, the paper straw to the plastic straw. AI is so Progressive. Our lives are finally going to be awesome, we won't have to be porn stars on Only fans anymore. All my prayers are answered.

   Reply
4. 

   @neilreid9005

   March 8, 2025 at 11:40 pm

   Hilarious when people think AI is going to take over the planet. Commercial grade AI is a loooooong way from any trick like that. It CAN be a good tool but you have to learn to use it well. It's absolutely retarded in some ways like mirroring, logic flaws you can't pull out, and you have to load one part of a large problem into it at a time. It blanks its memory after some sessions, and it really blows at analyzing spreadsheets with several thousand rows. It'll get better sooner than we think but it's largely over-hyped at the present.

   Reply
5. 

   @schubi42

   March 8, 2025 at 11:40 pm

   when LEGO ® realizes they had all the building blocks for AI for decades 😂

   Reply
6. 

   @hansschmidt148

   March 8, 2025 at 11:40 pm

   Predicting the next word(s) has absolutely nothing to do with intelligence. The term "AI" is misleading. Modern day LLMs are not smarter than a toaster. There is no reasoning, no substance and no consciousness there. It's just an illusion – it's believable for us because it appears to be intelligent. It is a nice toy and definitely an interesting tool for certain tasks, but it is nothing more than that. A tool.

   Reply
7. 

   @AaronKelly-m7n

   March 8, 2025 at 11:40 pm

   I'm not sure if he knows what he's talking about

   Reply
8. 

   @intheshell35ify

   March 8, 2025 at 11:40 pm

   A human's behavior is not derived entirely from it's particular history. We can generate spontaneous thought and growth in isolation. AI is bound to, and isolated by, it's particular history. It is nothing in isolation.

   Reply
9. 

   @EmeraldView

   March 8, 2025 at 11:40 pm

   Yeah stop watching and thumbs down this video after this clueless guy said all that these things are doing is predicting the next word. Wrong!

   Reply
10. 

    @rallisf1

    March 8, 2025 at 11:40 pm

    The main problem with AI is that it can perform well enough in tasks we usually enjoy like writing, coding, making art, etc. AI has also made leaps in material science and medicine, especially with alphafold, but It can't solve the financial crisis, the housing crisis, the global food crisis, or even do our chores.

    Is it because the process is flawed or perhaps AI doesn't have humanity's best interest in "mind"? I don't think so. Machine learning uses human generated content as its base dataset, thus it will be as flawed as the humans, the only difference being its compute can get it a few generations ahead. We don't need AI to solve most of our greatest problems, just better goals, less ego and greed.

    As far as daily tasks and chores go, most of them require many more senses than just reading (or listening) and remembering what a group of pixels in an image represents (which is how an AI can see). Even if we gave an AI powered robot all the senses we have, it takes an unimaginable number of iterations to learn the simplest of tasks. e.g. learning to make an omelete would require millions of eggs, if not more. So, unless we figure out a better way than the current transformers; yes AI has hit a wall.

    Reply
11. 

    @jonathanr72

    March 8, 2025 at 11:40 pm

    The problem I find is that it will try and insert and present woke dogma as fact.

    Reply
12. 

    @TheRadicalCentrist.1776

    March 8, 2025 at 11:40 pm

    What is the name of that equation? I have a master's in computer science, so I want to study it. I just don't know what it is. I appreciate the simple summary though!

    Reply
13. 

    @DEMOKRATEN-DEUTSCHLANDS

    March 8, 2025 at 11:40 pm

    The entire model is not matching to the phenomena it should describe. I know how it should be. I figured it out and realized an algorithm that works. BUT: K.Goedel proved already in early 1900 that algorithms will never become creative. Else they would solve problems beyond their programming foundations. And that's impossible. So: yes, AI is a hype only, it's possible to do something that actually works, but no: it'll never work independently from human creativity (including inventions).

    Reply
14. 

    @Derek-np7ke

    March 8, 2025 at 11:40 pm

    I think AGI is a horrible idea, we weren't exactly kind to the native Americans.

    Reply
15. 

    @markkukeskioja

    March 8, 2025 at 11:40 pm

    The 4 minute timer ruined the whole video for me, I can't concentrate on anything else while the timer is distracting me in the corner. I wanted to like this video

    Reply
16. 

    @AthenaNKnight

    March 8, 2025 at 11:40 pm

    3.9k dislikes

    Reply
17. 

    @raphaelvogt3215

    March 8, 2025 at 11:40 pm

    I like your example how you break the sentence into tolkins.

    Reply
18. 

    @edclune5357

    March 8, 2025 at 11:40 pm

    Slightly off your main topic, but I've been working on a friend generating some concept art using AI (Midjourney I think) and the problem I see is that the AI always seems to converge to a "standard" type. I ask for an 19th century single-family house, I get something with a sagging roof. I ask for an animated character, and without a lot of restrictions, it always looks like a Pixar character. I can't draw worth a darn, so I have to rely on something or someone to generate useful images for concepts I'm working on. Using AI doesn't seem to help with new interesting images but tends towards the popular ideas of the era. I've heard of the job of "Prompt Engineer" as a solution to this, but I have my doubts.

    Reply
19. 

    @Melanatedone

    March 8, 2025 at 11:40 pm

    The problem is in the approach. Now I can give you the solution, but why make it that easy? The adventure is in the journey, NOT the end of the book/movie.

    Wink, wink.

    Reply
20. 

    @WillowEpp

    March 8, 2025 at 11:40 pm

    The "clues" about the limitations of current model approaches have been known for years. Only credulous morons think any of this is a shock.

    Reply
21. 

    @azzzn-m8h

    March 8, 2025 at 11:40 pm

    what about manus ai

    Reply
22. 

    @MyUncleHere

    March 8, 2025 at 11:40 pm

    Apparently AI is not smarter, Grok or chat GPT is not smarter, only human getting more stupid the more they fascinated by its fake hype 😂

    Reply
23. 

    @theultimatereductionist7592

    March 8, 2025 at 11:40 pm

    ChatGPT is helpless at answering basic music theory questions. My brother has a doctorate in musical arts and composition. He gets very frustrated at the complete stupidity of ChatGPT and other AI programs at being unable to answer elementary questions such as "name this chord".
    I have a Math PhD. ChatGPT is unable to solve math problems in differential algebra or solve basic problems in Galois Theory.
    I have asked an AI program: "If Tom Copton is in Illinois in 1856, then where is Tom Copton's liver in 1856?" It cannot answer.

    Reply
24. 

    @pieeater1203

    March 8, 2025 at 11:40 pm

    hey thats linear algebra where i didnt expect it

    Reply
25. 

    @frank.sophia

    March 8, 2025 at 11:40 pm

    AI can take in all human information and process it meaningfully and practically.
    You wanna trust people instead?
    If there's anything current events should be teaching you it's that people suck.
    Sure AI is designed by us, but it can be programmed with our ideals.
    We have never met them and we are finite.

    Reply
26. 

    @sokak01

    March 8, 2025 at 11:40 pm

    This video is cooler than informative.

    Reply
27. 

    @atananaev

    March 8, 2025 at 11:40 pm

    Seems like the video contradicts the title "AI Has a Fatal Flaw—And Nobody Can Fix It". Maybe the flaw refers to single-shot models, but then the title still doesn't make sense because we have a solution.

    Reply
28. 

    @adfaklsdjf

    March 8, 2025 at 11:40 pm

    I downvoted because your arguments don't support your conclusion, and by the end of the video (17:17) where you're saying "it's no longer an 'if' but a 'when' question", you seem to be drawing the opposite conclusion from what the thumbnail, title, and beginning of the video presented. I think the title, thumbnail, and beginning of the video were designed to hook people who hate "AI" by telling them what they want to hear. Clickbait.

    Next token prediction alone probably has reached its limit or is very near its limit. A critical flaw in the message the video implied (until the end) is an assumption that "AI" == GPT next token prediction, as if no other methods will be developed. It's clear to me, at least, that there are many avenues to pursue from here that go beyond next-token-prediction.

    Aside from presenting an initial thesis that it was unable to truly support, the video was fine. Swap "AI" for "GPT" throughout, and I wouldn't disagree.

    Sidenote, whether or not what's been developed today constitutes "intelligence" or not is a semantic question, so I don't find it a compelling question. The reality is that the systems today have a set of capabilities and a set of flaws and limitations. Assuming systems continue to advance, developing new capabilities, they will continue to have flaws and limitations. Humans have flaws and limitations as well. Debating whether or not it's "intelligence" is a distraction, not a meaningful discussion. "AI" is the widely-established term we have for these systems, whether it's an accurate term or not.

    On the flawed terminology point, AI Explained—hands down the best channel covering AI developments—in a video he released on Patreon he made a very convincing argument that the term "AGI" is unsalvageable from a definition standpoint, because all 3 words are fundamentally flawed in this context.

    Reply
29. 

    @SunnyAquamarine2

    March 8, 2025 at 11:40 pm

    I think it's hilarious that nobody seems to consider the motivation to lie by those "pulling the strings" who are only in it for money. 😂

    Reply
30. 

    @Nootropics_become_limitless

    March 8, 2025 at 11:40 pm

    “There is insufficient data for a meaningful answer.”

    Reply
31. 

    @TheMrCougarful

    March 8, 2025 at 11:40 pm

    This video will not age well.

    Reply
32. 

    @MX-op7nf

    March 8, 2025 at 11:40 pm

    AGI would not cook you an egg for breakfast because it knows animal agriculture is unethical.

    Reply
33. 

    @Nandan.kella19

    March 8, 2025 at 11:40 pm

    Fix ya insecurities

    Reply
34. 

    @MrLargonaut

    March 8, 2025 at 11:40 pm

    While working in AI research and development, channels like these really help us understand what the public sees when they look at AI at many levels. These videos inform ways we can talk to people about AI, whether it's a doomsayer who lives for the apocalypse, people in the home AI maker space, or folks like you who're engaging with the news and boiling it down.

    Reply
35. 

    @goodtipa

    March 8, 2025 at 11:40 pm

    I'm confused. You started the video by saying it's doubtful that AI will get exponentially smarter, you proven this point over the course of the video and then you finished by "not if but when"?

    Reply
36. 

    @hugh261

    March 8, 2025 at 11:40 pm

    Lol, let AI figure it out.

    Reply
37. 

    @chess-jazz

    March 8, 2025 at 11:40 pm

    Very well explained, thank you so much!

    Reply
38. 

    @nerd3131

    March 8, 2025 at 11:40 pm

    just basically give a tldr so i understand you have smth promising bro

    Reply
39. 

    @VoxelMusic

    March 8, 2025 at 11:40 pm

    Maybe this script was ai generated cause it didnt make sense.

    Reply
40. 

    @Xhi_Clox

    March 8, 2025 at 11:40 pm

    i kinda get the vibe of veritasium video

    Reply
41. 

    @inflationking1271

    March 8, 2025 at 11:40 pm

    This is old and outdated information. Do your homework

    Reply
42. 

    @VexFeet

    March 8, 2025 at 11:40 pm

    it really comes down to the laziness of pajeets, constantly feeding the AI models non-curated AI generated content, leading to an autocannibalisation of the ai model.

    Reply
43. 

    @robinheyer708

    March 8, 2025 at 11:40 pm

    There's something very child-like about pumping more and more data into a model and then asking it if it is smart now. According to Epoch AI a model will be trained on everything humanity has produced by 2028. (And, of course, all the junk that AI has flung into existence in the meantime). There will have to be a miracle in the meantime if this is to produce intelligence. AI was released upon the world in possibly the worst time ever with the wars both physical and cultural going on coinciding with an all time low trust in government since the disease of unknown origins spread across the globe and has so far only served their creators' pockets.

    A GPT is a nice brainstorming tool but unfit for any prolonged operational work as anyone who knows their trade will confirm. Anyone losing their job now because of AI may have a couple of tough months ahead but after that they'll be almost as much in demand as before when most of the AI-driven start-ups crash because there's no miracle cure for hallucinating artificial "intelligence" or the human touch.

    Reply
44. 

    @chess-jazz

    March 8, 2025 at 11:40 pm

    The hello T-shirt made me smile ^^

    Reply
45. 

    @Illvester42

    March 8, 2025 at 11:40 pm

    The marriage of ai and quantum computing shall birth us our new God.

    Reply
46. 

    @Kraus-

    March 8, 2025 at 11:40 pm

    I am not at all concerned about "AI". Call the chatbots and pattern matching programs whatever you want, but without agency they're nothing more than fancy specialised tools.

    Reply
47. 

    @HEYVIET

    March 8, 2025 at 11:40 pm

    Enter Diffusion LLM’s… Each step will unveil another novel approach, we’re in pre-alpha ai right now

    Reply