593
A.I.’s un-learning problem: Researchers say it’s virtually impossible to make an A.I. model ‘forget’ the things it learns from private user data (finance.yahoo.com)
submitted 2 years ago by to c/technology@lemmy.world
206 comments fedilink hide all child comments
 

I'm rather curious to see how the EU's privacy laws are going to handle this.

(Original article is from Fortune, but Yahoo Finance doesn't have a paywall)

you are viewing a single comment's thread
view the rest of the comments
[–] 7 points 2 years ago (1 child)

Retraining the model is incredibly expensive. That basically means not training the model with any user data, even if it slips in accidentally, by someone sabotage the training data, or even with consent (since consent can be revoked).

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 18 points 2 years ago (2 children)

    consent cant be revoked, theyre not even trying to get consent.

    They seemingly all have a "use first then ask for forgiveness" approach which should come around to bite them in the ass

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 6 points 2 years ago (2 children)

    Anything else is going to bite US in the ass. Asking for consent kills any kind of open source development. It puts AI solely in the hands of like three companies. Our economy is going to be very AI focused in the future, they would literally own all of us.

    You aren't getting paid either way so we might as well all enjoy the fruits of humanities labor freely instead of been forced into a subscription model of it.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 2 points 2 years ago (1 child)

    Asking for consent doesn't kill open source development. Consent is the very reason we have licensed code. MIT, Apache, GPL3... And development is done and code is reused in accordance of those licenses.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 2 years ago

    Making llms requires a stupid amount of data, much more than what is found in the creative commons. Same goes for image gen. Unless you have been accumulating data since forever through tricking people when they sign up to your website or app, you can't train anything without scraping most of the data.

    It has nothing to do with licensing but the fact that there just isn't enough "free-use" data.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 2 points 2 years ago (1 child)

    Except it does not?

    For example: https://commonvoice.mozilla.org/

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 2 years ago

    "Most of the data used by large companies isn’t available to the majority of people. We think that stifles innovation."

    Yes crowd sourcing is a solution but is only really possible if you are able to reach many people like Mozilla can. They only have 20k of hours up to date. Tortoise needed 50k hours and was made by one guy who open sourced it. He would not have been able to build without scraping YouTube.

    Crowd sourcing also becomes much more complicated for llms or if you are making models in other language.

  • permalink
  • fedilink
  • source
  • parent