Close Menu
    Technik

    Anthropic behauptet, Destillationsangriffe im industriellen Maßstab durch DeepSeek, Moonshot AI und MiniMax identifiziert zu haben.

    Share.
    View 30 Comments

    30 Kommentare

    1. the_red_scimitar on

      And Google hilariously claimed this was happening to it’s AI, from many of the same players.

      All of them stole from all of us, then from each other. A perfect metaphor for the US situation.

    3. Speak_To_Wuk_Lamat on

      Behold the field in which I grow my fucks. Lay thine eyes upon it and thou shalt see that it is barren.

    4. JaggedMetalOs on

      „We can scrape whatever we like but you’re not allowed to scrape us!“ 

    6. Nothos927 on

      Companies that openly say their industry requires intellectual property infringement are shocked their industry partakes in intellectual property infringement.

    8. dezignguy on

      How is it an „attack“ if all they have to do is query Anthropic’s model? Isn’t that just their model leaking information it shouldn’t be?

    10. DarthJDP on

      how is this any different than the data scraping anthropic did? I hope these competitive models steal everything from silicon valleys models. Zero sympathy and I hope they annihilate the industry.

      Edit. now I know to use Deepseek, Moonnshot, and minimax to get the same or better performance of anthropic for a fraction of the cost.

    11. echoplex21 on

      Honestly still more ethical than them torrenting or ripping apart books and scanning them to train their own models.

    12. Binary101010 on

      This is what Cory Doctorow calls „adversarial interoperability,“ and what the tech giants don’t want you to know is that many of them were *built* on doing this to other sites, then lobbying Congress to pass laws to make it illegal for others to do it them.

    13. Deto on

      Instead of blocking them, if you really wanted to mess with competitors, it would be interesting to try and feed them bad results instead.

    16. Disgruntled-Cacti on

      “Distillation attack” lol.

      “You cant steal our stolen data!”

    18. aquarain on

      So he is saying the distillates performed the same service cheaper and faster. What a strange thing for a CEO to say.

    20. ddgconsultant on

      This raises serious questions about AI model security and intellectual property. The scale of these distillation attacks is concerning for the broader AI ecosystem.

    21. outofband on

      That they are spinning it as a matter of national security is fucking hilarious

    23. WWIIICannonFodder on

      Eat shit Anthropic. I hope the Chinese companies steal everything. All the data is stolen from humans anyways.

    24. FalconSixSix on

      So the guys who „train“ their models on other people’s work and say it’s completely fine suddenly have an issue when someone tries to train their own model on another model

    25. simplycycling on

      Wait, AI companies are using AI to close the gap with their competitors? What are they thinking??

    26. DarkSkyKnight on

      It’s hypocritical but the open models are really not that great. There’s a reason people pay hundreds per head for Claude Code even though it’s trivially easy to install a local model. Anthropic should have just said nothing because the open models aren’t even competing for the same customers.

    27. PunnyPandora on

      Dario Ratmodei at Misanthropic is a genuine clown. Even closedai has contributed more to open source. They at least released models… What has misanthropic done other than steal everything and then charge people for it? Their moat is nonexistent, and the rest of their shares will be bought up by google sooner or later. Good riddance.

    29. calm_hedgehog on

      Oh, when you do it it’s newest AI model training will replace all workers for low cost, give us moneyz. When others do it, it’s a distillation attack, huh.

    Leave A Reply