Lemdro.id
  • Communities
  • Create Post
  • heart
    Support Lemmy
  • search
    Search
  • Login
  • Sign Up
☆ Yσɠƚԋσʂ ☆@lemmy.ml to technology@hexbear.netEnglish · 11 months ago

AI models collapse when trained on recursively generated data

www.nature.com

external-link
message-square
22
fedilink
  • cross-posted to:
  • technology@lemmy.world
  • technology@lemmy.world
  • hackernews@lemmy.smeargle.fans
  • technology@lemmygrad.ml
  • machinelearning@lemmy.ml
71
external-link

AI models collapse when trained on recursively generated data

www.nature.com

☆ Yσɠƚԋσʂ ☆@lemmy.ml to technology@hexbear.netEnglish · 11 months ago
message-square
22
fedilink
  • cross-posted to:
  • technology@lemmy.world
  • technology@lemmy.world
  • hackernews@lemmy.smeargle.fans
  • technology@lemmygrad.ml
  • machinelearning@lemmy.ml
AI models collapse when trained on recursively generated data - Nature
www.nature.com
external-link
 Analysis shows that indiscriminately training generative artificial intelligence on real and generated content, usually done by scraping data from the Internet, can lead to a collapse in the ability of the models to generate diverse high-quality output.
alert-triangle
You must log in or register to comment.
  • KnilAdlez [none/use name]@hexbear.net
    link
    fedilink
    English
    arrow-up
    23
    ·
    edit-2
    11 months ago

    Of course they do, that’s not really a shocking statement. I do like this research though, actually looking at how the underlying data distribution is forgotten.

  • Bloobish [comrade/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    23
    ·
    11 months ago

    So more or less AI inbreeding?

    • Owl [he/him]@hexbear.net
      link
      fedilink
      English
      arrow-up
      10
      ·
      11 months ago

      More like AI prion disease.

      (There’s also a process where people merge AI models together, and models are merges of merges with a largely unknown lineage, and in this space there’s a growing inbreeding problem.)

    • ☆ Yσɠƚԋσʂ ☆@lemmy.mlOP
      link
      fedilink
      English
      arrow-up
      9
      ·
      11 months ago

      lol best description of what’s happening

    • ChaosMaterialist [he/him]@hexbear.net
      link
      fedilink
      English
      arrow-up
      3
      ·
      11 months ago

      hAIpsberg

  • crime [she/her, any]@hexbear.net
    link
    fedilink
    English
    arrow-up
    18
    ·
    11 months ago

    AI kessler syndrome bloomer

  • magnificentTOAD [none/use name]@hexbear.net
    link
    fedilink
    English
    arrow-up
    12
    ·
    11 months ago

    So do i…

  • viva_la_juche [they/them, any]@hexbear.net
    link
    fedilink
    English
    arrow-up
    9
    ·
    11 months ago

    sicko-pog

  • BurgerPunk [he/him, comrade/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    8
    ·
    edit-2
    11 months ago

    So basically the only way to have good “AI” LLMs is for it to be a mechanical turk? shocked-pikachu

    • TreadOnMe [none/use name]@hexbear.net
      link
      fedilink
      English
      arrow-up
      7
      ·
      11 months ago

      What if my robot was just a guy?

    • Bay_of_Piggies [he/him, comrade/them]@hexbear.net
      link
      fedilink
      English
      arrow-up
      5
      ·
      11 months ago

      Not really AI if they can’t learn from one another.

      • BurgerPunk [he/him, comrade/them]@hexbear.net
        link
        fedilink
        English
        arrow-up
        2
        ·
        11 months ago

        You mean LLMs aren’t really AI??? Double shocked-pikachu

  • BountifulEggnog [she/her]@hexbear.net
    link
    fedilink
    English
    arrow-up
    7
    ·
    11 months ago

    indiscriminate use of model-generated content

    Indiscriminate is the key word in this paper. No one trains this way. Synthetic data and filtering out bad data are already very important steps for training and will continue to stay that way. With proper filtering and evaluation, models trained on synthetic data do better then the ones before.

    This is not the end of ai, like so many wish it would be.

    • ☆ Yσɠƚԋσʂ ☆@lemmy.mlOP
      link
      fedilink
      English
      arrow-up
      12
      ·
      11 months ago

      I don’t think this is going to be the end of AI either, and the corpus of data before AI generated content became prevalent is also huge. So, I don’t think there’s really lack of training data. I personally think this is more interesting from the perspective of how these algorithms work in general. The fact that they end up collapsing when consuming their own content seems to indicate that the quality of content is fundamentally different from that generated by humans.

      • BountifulEggnog [she/her]@hexbear.net
        link
        fedilink
        English
        arrow-up
        5
        ·
        edit-2
        11 months ago

        Yea that’s completly fair, I think ai models in general to have lots of interesting characteristics that are very different from humans. I just see a lot of people taking conclusions from papers like this that aren’t justified.

        • ☆ Yσɠƚԋσʂ ☆@lemmy.mlOP
          link
          fedilink
          English
          arrow-up
          7
          ·
          11 months ago

          Very much agree, and I find the whole hatred of generative AI is largely misguided to begin with. It’s interesting technology that has useful applications. Most of the problems associated with it ultimately trace back to capitalism, as opposed to any inherent problem with LLMs themselves.

  • Owl [he/him]@hexbear.net
    link
    fedilink
    English
    arrow-up
    6
    ·
    11 months ago

    It’s obvious enough that this will happen if you cycle one model’s output through itself, but they looked at different types of models (LLMs, VAEs, and GMMs) and found the same collapse in all of them. I think that’s a big finding.

  • LocalOaf [they/them, she/her]@hexbear.net
    link
    fedilink
    English
    arrow-up
    6
    ·
    11 months ago

    Rampancy no-mouth-must-scream

    • fox [comrade/them]@hexbear.net
      link
      fedilink
      English
      arrow-up
      5
      ·
      11 months ago

      Rampancy if instead of becoming hyper intelligent beyond moral constraints you instead got dementia

      • LocalOaf [they/them, she/her]@hexbear.net
        link
        fedilink
        English
        arrow-up
        2
        ·
        11 months ago

        biden-nibble Look mack, I should be going, my wife Dr. Jill Cortana has to use the email

  • ChaosMaterialist [he/him]@hexbear.net
    link
    fedilink
    English
    arrow-up
    6
    ·
    edit-2
    11 months ago

    geordi-no Recursively Generated Data

    geordi-yes Aggression to the Mean

  • D61 [any]@hexbear.net
    link
    fedilink
    English
    arrow-up
    2
    ·
    11 months ago

    filled-square = The answer generated by the most sophisticated generative LLM’s mankind will ever generate.

  • Evilphd666 [he/him, comrade/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    1
    ·
    11 months ago

    deleted by creator

technology@hexbear.net

technology@hexbear.net

Subscribe from Remote Instance

Create a post
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !technology@hexbear.net

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

  • Ways to run Microsoft/Adobe and more on Linux
  • The Ultimate FOSS Guide For Android
  • Great libre software on Windows
  • Hey you, the lib still using Chrome. Read this post!

Rules:

  • 1. Obviously abide by the sitewide code of conduct. Bigotry will be met with an immediate ban
  • 2. This community is about technology. Offtopic is permitted as long as it is kept in the comment sections
  • 3. Although this is not /c/libre, FOSS related posting is tolerated, and even welcome in the case of effort posts
  • 4. We believe technology should be liberating. As such, avoid promoting proprietary and/or bourgeois technology
  • 5. Explanatory posts to correct the potential mistakes a comrade made in a post of their own are allowed, as long as they remain respectful
  • 6. No crypto (Bitcoin, NFT, etc.) speculation, unless it is purely informative and not too cringe
  • 7. Absolutely no tech bro shit. If you have a good opinion of Silicon Valley billionaires please manifest yourself so we can ban you.
Visibility: Public
globe

This community can be federated to other instances and be posted/commented in by their users.

  • 157 users / day
  • 731 users / week
  • 1.4K users / month
  • 2.98K users / 6 months
  • 14 local subscribers
  • 23.8K subscribers
  • 2.16K Posts
  • 24.8K Comments
  • Modlog
  • mods:
  • context [fae/faer, fae/faer]@hexbear.net
  • EmmaGoldman [she/her, comrade/them]@hexbear.net
  • SexUnderSocialism [she/her]@hexbear.net
  • gaycomputeruser [she/her]@hexbear.net
  • ZoomeristLeninist [they/them, she/her]@hexbear.net
  • UI: 0.19.8
  • BE: 0.19.11
  • Modlog
  • Legal
  • Instances
  • Docs
  • Code
  • join-lemmy.org