About the Authors

  • The Authors and Contributors of "Patent Docs" are patent attorneys and agents, many of whom hold doctorates in a diverse array of disciplines.
2018 Juristant Badge - MBHB_165
Juristat #4 Overall Rank

E-mail Newsletter

  • Enter your e-mail address below to receive the "Patent Docs" e-mail newsletter.

Contact the Docs


  • "Patent Docs" does not contain any legal advice whatsoever. This weblog is for informational purposes only, and its publication does not create an attorney-client relationship. In addition, nothing on "Patent Docs" constitutes a solicitation for business. This weblog is intended primarily for other attorneys. Moreover, "Patent Docs" is the personal weblog of the Authors; it is not edited by the Authors' employers or clients and, as such, no part of this weblog may be so attributed. All posts on "Patent Docs" should be double-checked for their accuracy and current applicability.
Juristat #8 Overall Rank


« Amgen Files Its Principal Brief in Amgen v. Sanofi | Main | "Zero Draft" of WHO CA+ Released »

February 06, 2023


ChatGPT cannot draft patent claims reciting novel and nonobvious features of an invention unless the user provides details of those features in the prompt. If you understand how generative language models work, this is self-evident. Otherwise, you may be among the breathless majority who seem to think that this technology is magic.

But if you provide details of novel and nonobvious features in the prompt then you are, as this article rightly points out, disclosing confidential information to OpenAI.

So, no, ChatGPT cannot assist an ethical patent attorney to draft patent applications (except, perhaps, for the background section).

Interesting...I think I wrote a paper about this in 2021. See "Limits of Using Artificial Intelligence and GPT-3 in Patent Prosecution"



The question "Can?" must be answered in the affirmative.

But that question is hardly the best question.

Others that come to mind:

- Most efficacious?
- Best combination of efficient and effective?
- Should?

Another consideration beyond that of inputs NOT being private, is that of any continual learning, especially of others. There may well be NO manner of tracking any specific input gained from that input to anyone else's generated output, so it is likely impossible to show derivation.

Also, the 'salve' of having a year to file once anything is shared to the public (given the utter lack of ability to trace derivation) would be COLD COMFORT if before that year was up someone else simply filed quicker. Given as there is no such thing as swearing behind, the fact of the matter is that someone else using the tool COULD benefit from your public sharing and even file THAT SAME DAY benefiting from your public sharing and beating you to a filing, thus making your own work** to be prior art against you.

**of course, you would never even be able to prove this was one's own work.

Also, I took a peek at Prof. Tu's article and simply saw an egregious error in the abstract that stopped me cold:

"The underlying basis for patent law is a quid pro quo: the inventor discloses how to make and use the invention and in exchange for that disclosure, the public gives the inventor an exclusive right to practice that invention, for a limited time"


This is NOT the bargain.

The patent right is NOT an affirmative act to practice anything - it is a negative right to keep others from practicing.

This is most easily (and quickly) grasped with the understanding of improvement patents that build on the inventions of others (and to those others, the negative right is OFTEN still in place - which means that one does NOT have an affirmative right to practice one's own (improvement) invention.

The comments to this entry are closed.

April 2024

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30