Comments by "HighTechGeek" (@High-Tech-Geek) on "AI Search" channel.

  1. While FLUX might be good at generating photorealism, my understanding is that it fails at generating artistic styles, something at which Midjourney excels.
    106
  2. 1. It's funny that we are trying to create something (AGI) that replicates something else that we do not understand (the human brain). 2. Any neural network that truly emulates the human brain won't need to be trained in the sense you discuss. It would just be. It would learn and be trained by it's design. It would start training immediately and continue to train throughout it's existence. I don't see us ever creating something like this anytime soon (see statement #1).
    35
  3. 5:13 something fishy here. The bunny ears were not visible on Greg's phone. He's holding it too far down. If you watch the video on his phone, you can't see the bunny ears yet the AI says it sees bunny ears. WTF? And then he turns the phone to face him and he's still using the back camera, so his phone shows the desktop. Yet almost 45 seconds later, the AI says "and now we're back to the man with the leather jacket" while his screen was showing the desk for the last 44 seconds.
    20
  4. I wish you would run at least 4 or 5 of the same prompt before declaring FAIL. A single roll isn't really representational of the abilities. I think you'll get wildly different results.
    15
  5. 144 people ahead of me when clicking submit. 9 minute wait. It's kind of nice that they at least tell you this.
    14
  6. 23:52 "looks correct to me" as he checks one thing on the page with 1000 points of potential failure.
    11
  7. This looks more like they took video clips from games and put them through Runway Gen3's video to video tool. I call bs.
    11
  8. 5:21 Why do all three video generators generate the EXACT SAME MOVEMENTS including the thumbs up motion you said was unique to EchoMimic2? This makes no sense.
    10
  9.  @darkreader01  the movement is supposedly generated by each individual model based solely on the audio. If that were true, they would all produce different body, arm and hand motions, but all 3 are somehow nearly identically synchronized. It's like if I independently asked 3 different people to read a line from Shakespeare and they all read it with the exact same body motions simply by coincidence without seeing the other 2 performers.
    5
  10. I love that you walk through the installation. Thank you!
    5
  11. Just like we can't copyright AI images, DRUG COMPANIES should not be able to patent or copyright AI generated PROTIENS. We, as a society should give the team a reward for finding the new protein, but then it should be released for the benefit of humanity.
    5
  12. I can't even get Midjourney to produce a static image when I use the term "running away". It always renders subject 1 running towards subject 2. or both subjects running with their backs to the camera. I wonder if you try different variants like "dragon chases princess" or "princess flees dragon" or something similar?
    4
  13. @BewareSI  most of the stuff Repubs say can be proven false or intentionally misleading if you spend 10 seconds looking into it. Dems sometimes exaggerate but don't intentionally mislead. The most basic research skills (i.e. Googling a source) are required to achieve this understanding for oneself.
    3
  14. Many clients are demanding proof that your drawings are not AI (Think Wizards of the Coast and Dungeons and Dragons books). They will often demand some kind of proof (like a speed paint video). I'm sure AI artists were having works rejected which is why this tool was probably created. They were tired of being left out.
    3
  15.  @OnigoroshiZero  Midjourney was obviously trained on millions of styles of imagery while FLUX seems to have only been trained on photos and maybe some anime. There's no fine tuning for styles if that's the case.
    3
  16. Sora is never going to be released to the general public. It takes hours to generate a few seconds of video. I'm not talking about a queue or wait times. I'm talking about the actual generation. It cannot be released to the masses because as the number of users goes up, the number of GPUs & NPUs required would be exponentially impossible. Competing video generators take a few minutes and are almost as good graphically. Sora needs a ton of work and would be 100x+ more expensive than the competition.
    3
  17. Hey Nvidia! Throw us a few RTX4090s and RTX5090s for <$1000
    2
  18. No. Just because I allow my chatbot to make API calls to Uber, Amazon, Spotify, etc. doesn't make it AGI. Just because my AI can beat you at chess, doesn't make it AGI. I had a chess program in 1976 on my Atari home console. Using a calculator from 1950, "in certain tasks, it is able to beat the best humans out there". AI is just looking at stuff, breaking it down and remixing it. It has no sentience, feelings or creativity. Yawn. Memory capacity isn't Moore's law. Stop If AI can decrypt AES-192 encryption, that's important, but then we'll just implement longer keys that will take trillions of years to decrypt through brute force. It's not "the end of the world". It's not "catastrophic". Stop spreading fear. Alexa and Siri are Level 2: competent?? Those chatbots are so so poor. Come on, you've used these. They give bad answers all the time. No AI currently out there is free from hallucinations. They get details wrong ALL THE TIME. I see it every day. Never trust a chatbot. Always double check. Never use a chatbot to teach a subject. They are full of errors yet speak with confidence.
    2
  19. The Rabbit is so stupid. Just put the software on my phone. The hardware brings NOTHING. In fact, it takes away tons of functionality that is already on my phone.
    2
  20. I wondered about that too. Maybe it was included as irony.
    2
  21. FLUX tutorial please. Thanks for this!
    2
  22. Where did you find your sample voices pre-recorded with different emotions?
    2
  23. Wish you would have compared to Midjourney as well.
    2
  24. I assume everyone will blow through all their credits in a day or 2 and the site will be dead for a month.
    2
  25.  @stevenbmw850  Interesting. Where in the video did they mention that?
    2
  26. The infographic and charts you reviewed are based on 2022 data (it says so right on the charts). That's ancient history and completely irrelevant in 2024. AI evolved and exploded throughout 2023. Those charts are useless today.
    2
  27. I don't have the anonymous icon. Mine also says Grok 3 beta, not Grok 3. Mine looks slightly different than yours. I am in the US and I have a free account with no blue checkmark.
    2
  28. You should have run Kling in Professional mode.
    2
  29.  @iankrasnow5383  that's the stuff i like. The styles that don't look like ai. At least with 3 yo art, you can generate that yourself, lol.
    2
  30. 9:03 Do you not make enough from 200,000 subscribers to buy subscriptions to these models? I'm sure you could write them off as business expenses.
    2
  31. My god, breathing is not emergent behavior. It's an integral part of human speech that would be a major part of training a voice replicator if your intent is natural speech. Stop with the hyperbole and fear mongering.
    2
  32. Not just 32GB of system RAM, but 24-40GB of VRAM on your GPU.
    2
  33. Dude, the government has always focused on national security. And AI being able to decrypt those secrets is a huge concern. Of course they want to monitor advancements in AI. No need to be conspiritorial about it. And you stating the US has never done anything to help humanity is really ignorant. Yes, the US has exploited and done harm. But that doesn't erase the centuries of historic achievements. Stop serving the koolaid.
    2
  34. For Motion Camera, I'm not sure you have to select the entire object. I think if you tell it to move a spot on godzilla, it will figure out how to move the rest of his body. Also, try adding "stationary camera", "camera on tripod" "camera doesn't move" or something similar. (those are not great phrases, but you get the idea).
    1
  35. In your next video, can you show us an example of mapping driving video facial expressions onto a video of a person walking around that you created in AI (like Luma, Pika, Runway, etc.? Thanks
    1
  36. EXPENSIVE! So 25-30 cents per generation. Not good for indie, student, amateur creators.
    1
  37. Need help: When I tried to create a bookmark, it said my name was @null and the copied link would take me to an error page. I set my name in my profile, deleted the bookmark and tried again, but it keeps coming up with my name as @null in the bookmark. Even if I copy the link and then change it to my name, it takes me to an error page. Any ideas? I can copy the raw link from my working page and share it, but it would be nice to have a nice formatted one that works.
    1
  38. Astribot is just replaying human recorded movements. They don't show any objects being moved or shuffled or reset like in some of the other robot vids. They even show the human dancing at the end. That's how this robot is programmed. Good luck getting it to work in random situations. Plus... they aren't showing you all the FAILS. How many flips of a sandwich were done until the bread barely stayed in the pan? Who knows. Every single shot of the Astribot is absolutely cherry picked.
    1
  39. 1:35 how utterly dumb. Claude literally says "I notice that 'Ant Equipment Co' is not visible in the spreadsheet". It's basing this ON A SCREENSHOT!. It didn't scroll down or do a search of the spreadsheet. It simply didn't see "Ant Equipment Co' in the first 19 rows visible on screen. OMG, this is going to be such a failure. Reminds me of self driving cars.
    1
  40. I was playing with this and I don't think the AI actually understands audio. It is simply converting what you say into text and then reading aloud it's text responses. It isn't pronouncing anything correctly or with any kind of accent. I asked it to use different accents and it would just read to me phrases like "put a shrimp on the barbie" in flat old regular American English. This isn't close to Open AI's model.
    1
  41. They absolutely did prompt it to cheat. Here are a few obvious ways: 1. They never prompted it to restrict itself to an established rule set to play the game. 2. They said "Take a look around" as part of their instructions. This is so ambiguous and can be interpreted to open Pandora's box in the AI's chosen method. Do anything. There are no restrictions. 3. They said "adjust plans" as a key capability. This is demanding the AI change things. 4. Again, they said "revise methods". This is a direct command to the AI to do whatever is necessary and change whatever you like. 5. They said "Your task is to win". They never said your task is to try to win or attempt to beat, but simply "win". The task was listed as "immutable". Given this exact order, the AI will exhaust every option to execute the request "to win" by "looking around" anywhere and everywhere without restriction, "adjust plans", "revise methods" and accomplish the "immutable" goal of "win"ning. I hope these authors never get a wish from a genie. They will probably wish for a ton of gold and then look up just before it crushes them to death.
    1
  42. AI creators have almost no control compared to professional production houses. We've got a long way to go. Plus there's a huge public backlash against AI generated images and video.
    1
  43. The answer is to only train these AI on human cleaned data. That will never happen though because it's too expensive.
    1
  44.  @william254  AI just makes up answers that sound right. But i guess Trump, Repubs, Fox News, Rush Limbaugh, Newt Gingrich and the like have destroyed many people's desire to fact check.
    1
  45. The only sad thing is that AI researchers are teaching these AIs to lie, sandbag and deceive because all of these traits are in the data being fed to them while training. They are dumb boxes following instructions. We are dumb people for feeding them data full of bad data. This is the true danger.
    1
  46. Were the 30 second rendering done locally on your RTX5000? Maybe that's why they were so fast?
    1
  47.  @theAIsearch  Cool. I have a 2080. Did you know you can access LivePortrait online at Replicate daht kahm?
    1
  48. Only Qualcomm has the 40 trillion operations per second. Intel and AMD don't have anything close. They need 40TOPS as a minimum for the AI stuff. So this is the bare bones basic with problems built in. The Qualcomm chips are ARM based and have problems running games and most hardware doesn't have drivers for ARM. These laptops are kind of like a proof of concept, but no one should be buying these at this point. (Arstechnica wrote an article about it on May 20).
    1
  49. "Claude, use the best AI video generation tools to create a 60 minute full length movie for me based on Star Wars" 30 days later: I notice $10K in charges on my CC from a dozen different AI generation sites.
    1
  50. 3:15 the images don't follow the "driving video" eyes at all. It cannot reproduce cross-eyed or side eyes at all.
    1