Random Thoughts and Observations on Data Science and Beyond

Wright Flyer and LLMs: Patent or not to Patent


A Flying Machine patent diagram, the 1903 Wright Flyer (National Air and Space Musuem, Washington D.C.), transformer architecture in the “Attention is All You Need” paper.

The Wright brothers patenting of their invention was arguably one of the reasons for the development of the aerospace industry.

A quick search on Google Patents and a check on GPT4 suggests that Large Language Models (LLMs) are not being patented. Why? If the LLM is truly as significant as the seminal developments like the Wright Flyer then why are they being guarded as trade secrets.

The Wright brothers invention was essentially a private and not a public-funded pursuit. So, it is not different from LLMs in this regard. An LLM patent application could help spur more innovation. A trade secret based approach is likely to be very limiting. What do you think?

Note-1: If the argument against patenting of LLMs is that they are more like weapons of mass destruction then they will need direct government involvement to regulate and intensively monitor.

Note-2: The 1903 Wright Flyer is housed at the National Air and Space Museum, Washington D.C. Better picture and a 3D view are available here. The Flying Machine diagram corresponds to a later patent by the Wright brothers. It is available here. “Attention is All You Need” is a seminal paper that spurred the development of LLMs. So, patenting isn’t essential. Often the model weights of even open source models are not available! However, in-principle, patenting requires adequate detail to be disclosed in order to allow for replication (after 20 years). Twenty years is a huge protection in this field.

Note-3: In the early 1900s the use of patents as a defensive weapon was not a common strategy as is often the case in recent times.

Aniruddha M Godbole is an AI & Data Science practitioner. He is a continuous learner. These are his personal views.