New York Times vs. OpenAI

ChatGPT vs. NY Times: The Lawsuit Doomed to Fail

Or: Why lawyers should try to understand Generative AI

Image by the author & Midjourney

Is the NY Times’ lawsuit against OpenAI some kind of clever PR stunt or are they serious?

Because that lawsuit seems to imply that large language models (LLMs) like the ones running behind ChatGPT are using the NY Times’ content directly in generating responses…

… which is a fundamental misunderstanding.

(and maybe, just maybe, the NY Times’ lawsuit has something to do with the fact that OpenAI just made a deal with multinational mass media company Axel Springer?)

The Misconception of AI

We’ve gone through three years of “WTF? AI can do THIS?!” and since late 2022 there is an unprecedented mainstream adaption of AI tools, but still many people don’t understand what’s behind the new magic toys.

Without knowing how AI works, it’s not a good idea to judge.

Language Models Are Not Replicating Content

The key misunderstanding in the NY Times lawsuit lies in the perception of how Large Language Models (LLMs) like ChatGPT use the data they are trained on.

The one thing I tell people right at the beginning of my workshops: AI models don’t simply store and regurgitate their training data (like specific articles from the NY Times). Instead, their training on vast arrays of text means that these models learn patterns and structures found in written language.

LLMs then use the learned patterns to generate new content that can resemble the style or structure of their training data but is not, by definition, a direct copy of it.

What is Generative AI And Can It Create Original Content?

LLMs, through their training, develop a statistical understanding of language. They exist in the form of a huge set of mathematical functions, that, when asked to generate text, are not…