{"id":5281,"date":"2024-11-27T15:39:56","date_gmt":"2024-11-27T15:39:56","guid":{"rendered":"https:\/\/www.patentclaimmaster.com\/blog\/?p=5281"},"modified":"2024-12-01T03:09:35","modified_gmt":"2024-12-01T03:09:35","slug":"patent-drafting-improvements-claimmaster-2024-5","status":"publish","type":"post","link":"https:\/\/www.patentclaimmaster.com\/blog\/patent-drafting-improvements-claimmaster-2024-5\/","title":{"rendered":"Patent Drafting Improvements in ClaimMaster 2024.5"},"content":{"rendered":"\n

ClaimMaster 2024.5 features significant improvements to our patent application drafting tools<\/a>. Note that GenAI features are available to our +Drafting<\/strong> add-on subscribers.<\/p>\n\n\n\n

Local document workspaces<\/h2>\n\n\n\n

This is a big one<\/strong>! ClaimMaster now lets you use local documents to improve the quality of GenAI output for patent application drafting. ClaimMaster utilizes a local database populated with text sections pulled from the selected documents to perform Retrieval Augmented Generation (RAG) <\/a>when generating output text based on your GPT\/LLM prompts.  With our RAG implementation, when you send any query to GPT\/LLM, ClaimMaster will first perform a semantic search of the local database that contains sections of your documents to find the most conceptually similar text snippets for your prompt and will then add those samples to the final prompt to LLM source. This approach dramatically reduces LLM “hallucinations” and improves the factual and stylistic quality of the output from the off-the-shelf AI models that will now reflect your preferred content and style.<\/p>\n\n\n\n

\n
\"ClaimMaster<\/a><\/figure>\n<\/figure>\n\n\n\n

As part of setting up a RAG within ClaimMaster, you will be able to specify various local “document workspaces” referencing specific documents\/directories, so that that GPT\/LLM prompts for different clients or technologies could be evaluated in their unique context by LLMs:<\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n

For more information on how to set up local document workspaces for your LLM prompts, please take a look at this step-by-step tutorial<\/a>.<\/p>\n\n\n\n

Improvements to figure descriptions generation with GPT and Llama vision models<\/h2>\n\n\n\n

ClaimMaster 2024.5 also allows you to pass multiple figures to OpenAI GPT vision models (i.e., gpt-4o or gpt-4o-mini) at the same time, so that you can have GPT write a description of several figures simultaneously. To do so, simply select multiple figures from a file using the page selection tool when sending a prompt to GPT\/LLM. <\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n

ClaimMaster will then insert all part numbers from the selected figures in the GPT prompt and will also transmit all selected figures to GPT for description generation. We’ve also updated the text of our default figure description prompts to accommodate multi-figure selections:<\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n

In addition, we can also pass figures to the new llama3.2-vision<\/a><\/strong> model that can be locally deployed with Ollama<\/a>. At this time, llama3.2-vision can reliably handle only one image per prompt. Otherwise, it is great addition to the local LLM library and can generate reasonable first-draft descriptions from the supplied figures. It’s still not as good as GPT, but also runs completely locally and is free, so it’s definitely worth trying out.<\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n

Improvements to Part Number Renumbering<\/h2>\n\n\n\n

We’ve also simplified and improved our Part Number Renumbering<\/strong> tool. Now, if you click on the “Show all parts identified in the Specification<\/strong>” checkbox in the tool, you’ll be able to view all of the part #s in the Specification sored in ascending order rather than grouped by the associated figure #s. This will let you to renumber only the selected part #s with the specified positive or negative step increment without renumbering related figures, allowing for a more finessed part renumbering operations.<\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n

For more information about other improvements in this release, click on the links below:<\/p>\n\n\n\n