GNU social JP
  • FAQ
  • Login
GNU social JPは日本のGNU socialサーバーです。
Usage/ToS/admin/test/Pleroma FE

  • Public

    • Public
    • Network
    • Groups
    • Featured
    • Popular
    • People

Notices by lain (lain@fediffusion.art), page 2

  1. Embed this notice
    lain (lain@fediffusion.art)'s status on Friday, 09-Jan-2026 07:59:41 JST lain lain
    i found that i can easily replace myself with claude and chatgpt, but it's not easy for other people to do so. there still seems to be some value to 'knowing what you're doing' and basic understanding. don't expect it to last too long.
    In conversation about 3 months ago from fediffusion.art permalink
  2. Embed this notice
    lain (lain@fediffusion.art)'s status on Tuesday, 06-Jan-2026 00:37:15 JST lain lain
    played with 5.2 codex medium. it's not bad but it does make mistakes. it's not 'hands off', but it asks you about what to do all the time so it should be alright.
    In conversation about 3 months ago from fediffusion.art permalink
  3. Embed this notice
    lain (lain@fediffusion.art)'s status on Thursday, 01-Jan-2026 02:07:41 JST lain lain
    in reply to
    • kaia
    @kaia @hannah pro-cat
    In conversation about 3 months ago from gnusocial.jp permalink
  4. Embed this notice
    lain (lain@fediffusion.art)'s status on Thursday, 01-Jan-2026 02:06:44 JST lain lain
    in reply to
    • kaia
    @kaia frankly, i see too many people moving around in cars who never even learned to cross the country on foot. this will lead to a complete loss of walking skills.
    In conversation about 3 months ago from fediffusion.art permalink
  5. Embed this notice
    lain (lain@fediffusion.art)'s status on Wednesday, 31-Dec-2025 02:00:31 JST lain lain
    new hobby: taking an open core project and asking chatgpt to implement the missing proprietary functionality
    In conversation about 3 months ago from fediffusion.art permalink
  6. Embed this notice
    lain (lain@fediffusion.art)'s status on Wednesday, 31-Dec-2025 02:00:29 JST lain lain
    in reply to
    • mischievoustomato
    @mischievoustomato so far perfectly
    In conversation about 3 months ago from fediffusion.art permalink
  7. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 29-Dec-2025 16:48:49 JST lain lain
    From an THE TRUTH ABOUT AI vid

    > The 'ELIZA effect' describes the human tendency to attribute intelligence and understanding to AI, even when its mechanisms are simple keyword matching.

    How long can people be in denial about this? Eliza was a very simple pattern matching system that tried to emulate the rather content-less psychonalytical therapy talk ("tell me about your father"). Current LLMs are nothing like it. You can give them tasks that everybody would have seen as requiring intelligence two minutes ago, and they can hold sensible conversations about nearly any topic.

    Saying that that isn't intelligence is like saying that actually, airplanes do not really fly, because flying in biological systems involves flapping wings, and airplanes don't do that. So it might look like flight but it really is something different.

    OK, i don't care about the word you use for that ability, but when you have a machine that can talk to you and work together with you to solve hard problems that would absolutely have required another human before, that's a lot more than a markov chain or a "you mentioned crying. how does that make you feel?" text matcher.
    In conversation about 3 months ago from fediffusion.art permalink
  8. Embed this notice
    lain (lain@fediffusion.art)'s status on Saturday, 27-Dec-2025 03:44:17 JST lain lain
    in reply to
    • kaia
    @kaia it's sandboxed to the folder you run it in
    In conversation about 3 months ago from fediffusion.art permalink
  9. Embed this notice
    lain (lain@fediffusion.art)'s status on Saturday, 27-Dec-2025 03:44:04 JST lain lain
    • kaia
    • Terry Hendrix II 🏹
    @thendrix @kaia i really wonder what google, apple and AMD are working on right now
    In conversation about 3 months ago from fediffusion.art permalink
  10. Embed this notice
    lain (lain@fediffusion.art)'s status on Saturday, 27-Dec-2025 03:15:20 JST lain lain
    in reply to
    • kaia
    @kaia it has a command line application that you chat with and it works on the local repo for you, or you let it do online on GitHub. I only use the local repo version.
    In conversation about 3 months ago from fediffusion.art permalink
  11. Embed this notice
    lain (lain@fediffusion.art)'s status on Saturday, 27-Dec-2025 02:39:46 JST lain lain
    after a few more days with ChatGPT 5.2 on xhigh, this really is qualitatively different from the coding models before it. It can work autonomously for at least a day, and deliver high quality code.
    In conversation about 3 months ago from fediffusion.art permalink
  12. Embed this notice
    lain (lain@fediffusion.art)'s status on Wednesday, 24-Dec-2025 02:15:06 JST lain lain
    using coding agents to decompile nintendo 64 games

    https://blog.chrislewis.au/using-coding-agents-to-decompile-nintendo-64-games/
    In conversation about 3 months ago from fediffusion.art permalink

    Attachments


    1. https://fediffusion.art/media/93bf98e73b6d94c87df8533eeeae272a94e39724024a000241e8be3a9e25a9d0.png
    2. Domain not in remote thumbnail source whitelist: blog.chrislewis.au
      Using Coding Agents to Decompile Nintendo 64 Games
      A look at where coding agents help (or don’t) in the matching decompilation of Snowboard Kids 2 on the Nintendo 64.
  13. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 22-Dec-2025 06:26:40 JST lain lain
    in reply to
    • :suya:
    @newt sure, but that's your employer's problem. maybe their code is legitimately so secret that it can't touch anything else, that's alright. but if it isn't, keeping it away from LLMs will make it less secure and less efficient, which can't help in the long run.
    In conversation about 3 months ago from gnusocial.jp permalink
  14. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 22-Dec-2025 06:26:00 JST lain lain
    it's getting to the point where it's unconscionable to not use LLM assistants for programming
    In conversation about 3 months ago from fediffusion.art permalink
  15. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 22-Dec-2025 06:15:47 JST lain lain
    with 5.2, it really has reached the "it's over" level. it doesn't spiral out of control and can continue a task over several days with minimal input.
    In conversation about 3 months ago from fediffusion.art permalink
  16. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 22-Dec-2025 06:15:46 JST lain lain
    in reply to
    • mischievoustomato
    @mischievoustomato people who don't use it really have no idea. they post their 'too many fingers' jokes and some random instance from 3 years ago when chatgpt 1 wrote fizzbuzz wrong.
    In conversation about 3 months ago from fediffusion.art permalink
  17. Embed this notice
    lain (lain@fediffusion.art)'s status on Monday, 22-Dec-2025 06:15:44 JST lain lain
    in reply to
    • mischievoustomato
    @mischievoustomato i feel like z image turbo is halfway there, will be fun when they release the undistilled model
    In conversation about 3 months ago from fediffusion.art permalink
  18. Embed this notice
    lain (lain@fediffusion.art)'s status on Friday, 19-Dec-2025 15:45:53 JST lain lain
    https://github.com/DGoettlich/history-llms

    cool idea, training LLMs only on data that's older than 100 years
    In conversation about 3 months ago from fediffusion.art permalink

    Attachments

    1. Domain not in remote thumbnail source whitelist: opengraph.githubassets.com
      GitHub - DGoettlich/history-llms: Information hub for our project training the largest possible historical LLMs.
      Information hub for our project training the largest possible historical LLMs. - DGoettlich/history-llms
  19. Embed this notice
    lain (lain@fediffusion.art)'s status on Thursday, 18-Dec-2025 16:33:46 JST lain lain
    Meta has released SAM (Segment Anything) for audio. This essentially means that you can take an audio source and split it up into the different audio sources it's made up from. Want to do a fan dub? remove all the voices but keep the sound effects and music intact. Want to stream a game with copyrighted music? Just remove it while keeping the rest.

    https://ai.meta.com/samaudio/
    In conversation about 4 months ago from fediffusion.art permalink

    Attachments

    1. Domain not in remote thumbnail source whitelist: scontent-bos5-1.xx.fbcdn.net
      SAM Audio
      With SAM Audio, you can use simple text prompts to accurately separate any sound from any audio or audio-visual source.
  20. Embed this notice
    lain (lain@fediffusion.art)'s status on Tuesday, 16-Dec-2025 06:20:56 JST lain lain
    in reply to
    • mischievoustomato
    @mischievoustomato weirdly, the training tended to generate nude pictures after a while, probably becuase she's nearly nude in so many pictures, so in contrast to pictures without the trigger word, it'll keep undressing.
    In conversation about 4 months ago from fediffusion.art permalink
  • After
  • Before

User actions

    lain

    lain

    Tags
    • (None)

    Following 0

      Followers 0

        Groups 0

          Statistics

          User ID
          203505
          Member since
          23 Oct 2023
          Notices
          136
          Daily average
          0

          Feeds

          • Atom
          • Help
          • About
          • FAQ
          • TOS
          • Privacy
          • Source
          • Version
          • Contact

          GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.

          Creative Commons Attribution 3.0 All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.