OpenAI opens a conversational model to the public, and every prompt typed into it leaves the company

2022-11-30Industry

OpenAI, "Introducing ChatGPT," openai.com (Nov. 30, 2022)

OpenAI released ChatGPT, described as a model that "interacts in a conversational way," as a free research preview open to the public. The announcement states that the model is fine-tuned from a model in the GPT-3.5 series that finished training in early 2022, that it is a sibling model to InstructGPT, and that "We trained this model using Reinforcement Learning from Human Feedback (RLHF)," with human trainers writing conversations in both the user and assistant roles and ranking model responses to build a reward model. OpenAI states that it is releasing the model "to get users' feedback and learn about its strengths and weaknesses" and that "During the research preview, usage of ChatGPT is free." The stated limitations are that "ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers," that it "is sensitive to tweaks to the input phrasing," that it "is often excessively verbose and overuses certain phrases," that it guesses at the user's intent rather than asking clarifying questions, and that it may respond to harmful instructions or exhibit biased behavior. Users are asked to report problematic outputs through the interface, and a feedback contest offers credits for the application programming interface. The announcement does not describe how conversation data is stored or used beyond the feedback mechanism.

What it changesThe patent problem arrives the first time an engineer pastes an unfiled invention into the prompt box to get help writing it up. The text goes to a third party under OpenAI's consumer terms rather than a negotiated confidentiality agreement, into a product OpenAI says it released to learn from its users. Whether that is a public disclosure under Section 102 of Title 35 of the United States Code is an argument no one wants to have, and the absolute-novelty jurisdictions give no grace period in which to have it. The in-house action is a written policy, issued now, that unfiled inventions, source code and anything marked confidential do not go into a public model, with an approved internal tool or an enterprise agreement carrying no-training and confidentiality terms as the alternative. The second consequence is slower: a model that writes plausible technical prose at this volume will fill the prior art with machine-generated descriptions, and examiners will cite them, so file on what matters sooner than planned.

More on Prior art, public disclosure and obviousness · Trade secrets and open release

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.