ChatGPT owner says it’s ‘impossible’ to train AI solely in the public domain; to understand

There is no way to train a fairly complex language model with only publicly available materials. This is stated by OpenAI, the company responsible for the ChatGPT artificial intelligence chatbot.

According to the newspaper, the company used this argument as a way to defend itself in front of the UK parliament. Telegram. He appealed to the territory’s legislature in an attempt to avoid legal action for using copyrighted material.

According to the company, platforms such as GPT-4, which works on both ChatGPT and Microsoft Copilot, would have great difficulties in terms of education without such studies.

OpenAI claims copyright coverage today “virtually any human expression”including books, internet texts, photographs, codes and government documents.

“Restricting educational data to public domain books and illustrations created more than a century ago might be an interesting experiment, but Cannot ensure AI systems meet the needs of today’s citizens“, says the company.

In practice, this means that chatbots can only be used if materials that are publicly available or duly authorized by their owners are used. would be more limited in terms of content creation and general information.

OpenAI takes over writers’ and newspapers’ processes

OpenAI’s appeal to the government is also a form of protection against lawsuits the company currently faces. Lawsuits regarding the use of copyrighted materials to power artificial intelligence such as ChatGPT It started in 2023 and should be intensified this year.

In July, artists and writers took legal action against Meta and OpenAI. The accusation is the same: The language models were fed copyrighted works without permission from those responsible.

As a result, AI not only becomes a subject matter expert, but also has the ability to create content that replicates the visual or writing style of the author. In a note at the time, Microsoft confirmed: will protect consumers accused of copyright infringement For use by Copilot.

In a subsequent lawsuit filed in December 2023, The New York Times also sued OpenAI. “Illegal use” of your materials to train AIs without paying any financial compensation.