United StatesProfessional services2023 – ongoingGunderson Dettmer
ChatGD in-house generative AI app
Nearly half of Gunderson Dettmer’s staff tried its in-house AI app within four months
Problem
Gunderson Dettmer, an international law firm based in Silicon Valley, had more than 400 lawyers. It wanted them to learn to use large language models on client documents while keeping confidential material out of public chatbots.
Approach
Gunderson Dettmer launched ChatGD in August 2023 after its legal engineering team built it with enterprise OpenAI models on Microsoft Azure. Lawyers could provide agreements as reference material through retrieval-augmented generation, and had to complete training before using the app. In December 2023, the firm added routing to send simple retrieval tasks to cheaper models and long summaries to a GPT-4 model with a larger context window. It also planned to connect the app to its document management system.
Results
Takeaway
The firm expected to keep operating costs low by hosting an open-source embedding model and assigning simpler tasks to cheaper models, excluding internal engineering costs. Its innovation chief treated the project as a learning exercise and said reliable higher-value legal work needed more engineering.
Categories
- Industry
- Professional services
- Business function
- Internal work and knowledge managementFinance, risk and legal
- Innovation type
- Organization and ways of workingProcess innovation
- AI technology
- Generative AI
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Gunderson Dettmer launched ChatGD as its first in-house AI tool after years of using outside products. The app used secure enterprise OpenAI models on Microsoft Azure and retrieval-augmented generation, with much of its software drawn from open-source components. The firm set internal AI guidelines to protect client confidentiality and work quality.
Legal technology reporter Bob Ambrogi reported that Gunderson Dettmer, an international firm with more than 400 lawyers, launched ChatGD. He described it as apparently the first US-based law firm to build its own internal generative AI tool. He explained how the app retrieved reference material and why the firm chose to build it.
Gunderson Dettmer told LawSites that nearly half its staff had tried ChatGD in its first four months and submitted more than 9,000 prompts. Training was mandatory, and more than half the staff attended live. The firm projected annual costs below $10,000, excluding internal engineering, by hosting its own embedding model and using cheaper models. New routing features assigned prompts to models by task.
Full details are in the original sources.