• 0 Posts
  • 2.43K Comments
Joined 2 years ago
cake
Cake day: August 27th, 2023

help-circle

  • Text generation is the least you can do. You can still fire up Photoshop and feed in a half-finished image. Diffusion turns whatever you have into whatever you describe. If it does decent scratches on metal, but won’t put them exactly where you want, then select them and move them, and the robot will smooth it over.

    The very first article I read about Stable Diffusion, three years ago, had the author doodling mountains and flipping a spaceship. All the image-to-video stuff demands you provide art, as an input. Prompts alone are just a tech demo gone feral.



  • Software has won.

    This was inevitable once ports looked the same and ran the same. Doubling your customer base, without developing the whole game twice? Obvious choice for any third party. First-party developers have taken longer, because their parent companies primarily own them to promote a hardware business. Microsoft’s hardware business has become vestigial. It always was, to some extent; the Xbox project was a 1990s scheme to PC-ify the console market. It worked.

    Consoles don’t exist anymore. Do you want the green AMD laptop, or the blue AMD laptop? Even Nintendo rebadged an Android tablet. You can release some crazy new hardware unlike anything else, but the only third-party games will be multiplatform hits that run like garbage. Like on early PS3. The Helldivers 2 PSN fiasco sure looks like Sony found out how profitable they’d be as just another publisher and the answer scared the shit out of them. Without that service, they don’t have a platform, anymore. They sell a popular model of an IBM compatible. Asterisk on the compatible.

    Nintendo can get away with that shit forever, because they own Pokemon. I don’t know how much longer you can cosplay that sort of first-party importance, on the strength of Horizon and… Death Stranding.


  • Nobody can “get” exclusives. They don’t exist anymore. There’s first-party games - there’s games the first party funded into existence - and everything else runs on whatever customers have.

    Developers want to sell games… to people. Hardware is an obstacle. They don’t want to care which color of deliberately incompatible generic computer you bought. Shipping five near-identical versions of the same damn game is a button they push in Unreal 5, and that’s why they put up with Unreal 5.



  • Right, should say deep neural networks. Perceptrons hit a brick wall because there’s some problems they cannot handle. Multi-layer networks stalled because nobody went ‘what if we just pretend there’s a gradient?’ until twenty-goddamn-twelve.

    Broad applications will emerge and succeed. LLMs kinda-sorta-almost work for nearly anything. What current grifters have proven is that billions of dollars won’t overcome fundamental problems in network design. “What’s the next word?” is simply the wrong question, for a combination chatbot / editor / search engine / code generator / puzzle solver / chess engine / air fryer. But it’s obviously possible for one program to do all those things. (Assuming you place your frozen shrimp directly atop the video card.) Developing that program will closely resemble efforts to uplift LLMs. We’re just never gonna get there from LLMs specifically.



  • Neural networks will inevitably be a big deal for a wide variety of industries.

    LLMs are the wrong approach to basically all of them.

    There’s five decades of what-ifs, waiting to be defictionalized, now that we can actually do neural networks. Training them became practical, and ‘just train more’ was proven effective. Immense scale is useful but not necessary.

    But all hype has been forced into spicy autocomplete and a denoiser, and only the denoiser is doing the witchcraft people want.