3 ms·
The issue with bigger LLM or even MMMs is that the bigger they are the more they are cramming and regurgitating the training data, and that opens up to lawsuits
by nojvek 3y ago
The issue with bigger LLM or even MMMs is that the bigger they are the more they are cramming and regurgitating the training data, and that opens up to lawsuits.
Making NNs generalize the way humans do it is still a hard problem.
- lewhoo 3y agoIs this indeed established ? Could you provide a link or three ?
- nojvek 3y agohttps://avidml.org/blog/llm-privacy-plaigarism/ https://avidml.org/blog/llm-privacy-plaigarism/ https://moreisdifferent.blog/p/weird-properties-of-large-language https://moreisdifferent.blog/p/weird-properties-of-large-lan... https://www.sarus.tech/post/llms-have-become-excellent-at-memorizing-its-a-problem-for-privacy https://www.sarus.tech/post/llms-have-become-excellent-at-me...
- lewhoo 3y agoThank you. That's certainly interesting.