#MerzSays: Authors vs. AI?

There are currently several lawsuits between some prominent authors and OpenAI and other companies that train their text models on literature.

The authors claim that they were not given the opportunity to opt out of the datasets or to be compensated for their participation.

My 2ct.

For the sake of fairness, the opt-out should definitely be available to authors. Better would be a way of rewarding authors for used works.

But… Their accusations of theft reflect their incompetence in how the system works. During training, LLM-models read all the texts and understand how the language works. After that, the model itself doesn’t contain any works, only vectors in representation space. Then the generated texts will never plagiarise, LLM will create new texts (as human literature always works).

Or should we ban libraries and book sharing to help authors to get their 1:1 rewards for their sold copies instead of 1 book to x readers?

Side note: I would love if libraries in Germany would get my book — I don’t want 1:1 rewards (1 book : 1 reader), I want my knowledge and experience to be spread across the country.

Authors’ texts in datasets belong to human knowledge and experience of AI, i.e. knowledge of future AI-driven solutions, i.e. the future of digital culture. Opting out makes authors’ works obsolete for the digital future of AI-driven systems that will be as common as electricity in every home. But fair and respectful treatment, compassion towards creators, should be established in every area of our world — including the preparation of training datasets.