Integration with Generative AI/LLMs

We are excited to add new generative AI/LLM capabilities in our patent application drafting tools and Office Action analysis tools. In particular, +Drafting and +Shells users now have access to GPT services, either directly from OpenAI or Microsoft Azure OpenAI. GPT is an advanced large language model for natural language processing that can generate coherent and contextually relevant responses to given prompts, making it a very helpful tool for patent application drafting and summarizing Office Actions.

Using Generative AI (GPT) for Patent Drafting

Patent drafters stand to save time and money by being able to swiftly generate extensive segments of patent applications using GPT technology. However, we realize that the quality of the generated text depends a lot on the prompt itself and the “creativity/randomness” of GPT’s response. Therefore, we let you edit, store and invoke the best GPT prompts on demand, as well as set the “creativity” level (i.e., “temperature”) for each result. Some of the helpful tasks you can ask GPT to perform include:

  • Explaining or defining a technical term
  • Re-writing a text portion more eloquently
  • Writing a paragraph or two explaining the state of problems with the current technology
  • Generating sample data tables
  • Drafting a section of the application based on the passed-in subject matter, such as title, background, etc.

ClaimMaster also lets you enrich your stored GPT prompts with selected text and other information that ClaimMaster extracts for you from the document, such as part names/terms, acronyms, claim terms and limitations, and more. You can also apply GPT transformations on the results produced by previous GPT prompts to run multiple text transformations in series. These new generative AI capabilities nicely complement other ClaimMaster patent application drafting tools that are deterministic in nature for situations where you need to match the original claim language in document sections.

To access the GPT, click on the Patent Drafting Tools menu, then select Generate Text with GPT tool. Next, specify the prompt template and any data to populate the prompt before it is sent to GPT. You can also edit the prompt directly in the preview window. Next, click on the “Send prompt to GPT“, which will send your query to GPT.

When GPT returns the result, you can review/edit the generated text, reuse another template prompt on it, and finally insert it into your document.

Summarizing Office Actions with GPT

You can also use GPT to summarize contents of Office Actions. To do so, in the Office Action analysis tool, switch to OA Text Summaries tab, then “Summary by OpenAI GPT” tab, then click on the Summarize OA with GPT button to direct GPT to generate a brief summary of the Office Action using your preferred “randomness/creativity” settings. You’ll be able to insert this summary in the OA shell as well with the [OASUMMARY_GPT] replacement field.

Available GPT Service Endpoints and Models

ClaimMaster integrates with OpenAI and Azure OpenAI services. Both integrations are currently free for +Drafting and +Shells users (this may change in the future). And the default gpt-3.5-turbo model is good enough for most cases. However, if you have your own account with OpenAI or Azure OpenAI, ClaimMaster will let you pass your API account information for completing GPT prompts (you’ll incur your own costs for the generated text). In particular, if you use Azure OpenAI, you may find this option beneficial as you’ll be able to use proprietary/private GPT models, more advanced GPT model versions, as well as set the max limit on the tokens to generate more text than allowed by ClaimMaster’s default settings.

Confidential Data Security and Factual Correctness

Of course, we understand numerous concerns that practitioners have about transmitting confidential information to generative AI applications, like OpenAI GPT. While you should determine whether OpenAI’s data usage policies and/or Microsoft Azure OpenAI policies suit your needs, we try to minimize the risks as follows:

  • First, in all cases, ClaimMaster will never contact or send a prompt to either OpenAI or Azure OpenAI GPT service without your explicit command from the GUI to do so (i.e., clicking a ‘Submit’ button in ClaimMaster window). You always see the exact text of each prompt before it is sent to GPT. As a result, you control exactly what data is included in the prompts sent to GPT from your Word document, if any.
  • Second, we integrate with OpenAI’s API gpt-3.5-turbo model, which ensures that OpenAI and Microsoft Azure will not train their models on submitted information according to OpenAI policies and Microsoft Azure OpenAI policies, unless we opt-in to share data (we do not and will never do). Note that according to OpenAI, any data sent through the API may be retained for abuse and misuse monitoring purposes for a maximum of 30 days, after which it will be deleted (unless otherwise required by law). In addition, OpenAI website’s states that its servers are currently located in the US. Microsoft Azure has similar provisions.
  • Third, for even more security, you can point ClaimMaster to your private GPT models, such as the ones hosted by Microsoft Azure OpenAI. In particular, you may already have a data service agreement in place with Microsoft, reducing data security and privacy risks. Microsoft makes a number of guarantees that your GPT prompts (inputs) and completions (outputs), your embeddings, and your training data are NOT available to other customers and OpenAI, are NOT used to improve OpenAI models., are NOT used to improve any Microsoft or 3rd party products or services, and are NOT used for automatically improving Azure OpenAI models for your use in your resource.
  • Finally, if you’d rather disallow access to GPT for your employees, we can also provide you with a ClaimMaster version that has GPT integration completely disabled.

It also goes without saying that you should always review GPT responses for factual correctness, especially if you are using GPT prompts with settings for higher response randomness/creativity. That said, we’ve had good results with requests to explain technology, define terms, rewrite sentences, and so on, when “creativity” level is set to a medium value.

For more information about this feature, check out the step-by-step tutorial and also the Online Manual.

Here’s more info on how to interface to custom GTP models/endpoints.

Click on the links below to read more detailed posts about other improvements in ClaimMaster 2023.5: