from [[Azure Tech Day March 2024 Edition]]
Talking with Tech Leads about how to use OpenAI in your company
- DX Planning Team Manager, System Planning Department, JAC Recruitment Co.
- Junya Tsuruta
- Executive Officer, General Manager, Information Systems Division, Information Systems Headquarters, Cybozu, Inc.
- Shuichi Suzuki
- General Manager, Intelligent Cloud Sales & Marketing Division, Corporate Solutions Business Group, Microsoft Japan, Ltd.
- Yasushi Kosugi
-
Emergence of ChatGPT and company reactions
- With the advent of ChatGPT in November 2022, the importance of early utilization was recognized.
- Mr. Tsuruta of JAC Recruitment expected the regulations to come into effect early and thought it was important to use them up as soon as possible.
- Mr. Suzuki of Cybozu hoped that the ultimate system could be created by including all internal information.
- Expectation that communication in natural language will create the ultimate system.
- We thought that by including all the information within the company, we would be able to track the progress of the project and the movement within the company.
- With the advent of ChatGPT in November 2022, the importance of early utilization was recognized.
-
Status of Efforts by Both Companies
- Both companies are in the process of implementing a search function that leverages their internal knowledge base. Feedback was incorporated and tuning of prompts proved to be important.
- We are in the process of implementing a search function that leverages our in-house knowledge base.
- A system was established to incorporate feedback and determine whether a response was good or bad.
- Cybozu Internal Initiatives
- Developed chatbots that learn internal institutional information, rules, and knowledge.
- The goal is to create a system that will help employees solve problems when they arise.
- The POC has been in place for nearly a year and is currently in the company-wide rollout phase.
- Tsuruta: Tuning of prompts proved to be important for improving accuracy.
- Bad answers are misinterpreted, good answers are enhanced and exceed expectations.
- Suzuki: AI Accuracy Improvement and Feedback
- Employees provide feedback on how good or bad the answers are, which is used to improve the system.
- We also devised a way to provide two answers to a single question and let people decide which is more appropriate.
- There is a gap between developers’ intentions and users’ expectations
- Both companies are in the process of implementing a search function that leverages their internal knowledge base. Feedback was incorporated and tuning of prompts proved to be important.
-
How to deal with AI
- Tsuruta: It is important to understand that AI is not perfect like humans, and to recognize that it has its own habits and to deal with them.
- Like human new hires, AI should be viewed as having room for growth.
- Suzuki: By giving them names and personalizing them, such as “Exciting-kun,” communication with employees will become smoother.
- Tsuruta: It is important to understand that AI is not perfect like humans, and to recognize that it has its own habits and to deal with them.
-
JAC Recruitment’s 4-step approach
- Level 1
- Effective use of internal information
- Internal Search
- Easy to experience value
- Effective use of internal information
- Level 2
- Automation of routine operations
- Middle and back office BPR
- Efficiency of internal work is less likely to cause problems outside the company if it fails.
- Automation of routine operations
- Level 3
- Increased consultant time productivity
- Consultant Assistant
- This is because the presenter’s industry is “a type of business model where the consultant’s work efficiency is the bottleneck.ビジネスインパクトがあるということ
- Increased consultant time productivity
- Level 4
- Creating CX (Customer Experience) using AI
- Level 1
-
Recruitment of human resources and formulation of guidelines
- AI engineers are scarce, accounting for only 5-6% of the total, and securing them is a challenge.
- Companies have begun to formulate generation AI guidelines, and the handling of information is being discussed.
-
Requests to Microsoft
- Efforts to unify guidelines and improve the reliability of information
- Copilot Copyright Commitment
- Review of Azure OpenAI Service pricing
- Some call for seamless integration to avoid fragmentation of services
In general, the speakers’ view was that, while not expecting too much, it is important to steadily expand the scope of utilization, share knowledge among companies, and promote the practical application of AI from Japan.
memo
- AI is sensitive to public relations because it is directly related to IR
- A story about how MS was told not to give the number of AI installations in each country.
- be regulated
- It takes from within the company.
This page is auto-translated from /nishio/テックリードと語る社内におけるOpenAI活用術 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.