The Balance Between Confidentiality and Productivity Improvement

- The ability to consult AI for OSS code will increase productivity. - OSS products progress faster and become more competitive. - Information that should be confidential may be limited to a portion of the entire product. - Companies need to weigh the benefits of source code [[confidentiality]] versus increased [[productivity]].

voluntas It is very good that we, a company whose products are mostly OSS, can have ChatGPT analyze our code as we like. I didn’t think that the advantage of being OSS would appear in such a place. voluntas This seriously is a benefit because employees and people helping me can consult ChatGPT for code for their products (OSS) without thinking twice. I think it’s a great benefit. Of course, if something goes wrong, error messages can be pasted directly into ChatGPT for consultation.

nishio Ah, if there is a big difference in productivity based on whether or not the source code can be read and consulted by an AI, than products that cannot have their source code read by an AI If there is a big difference in productivity between whether or not the source code can be read and consulted by an AI, will the OSS that can read the source code progress faster than the products that can’t? nishio I think most companies in the world have roughly fallen back on “source code is confidential, so you can’t let it out of the company”, but how much of it should be read by an AI? When you weigh how much of a productivity benefit there is in letting AI read it against how much risk there is in letting AI read it, the “secrets that should still be kept secret” are only part of the whole product.

An example of the development I was writing about the day before.

nishio Today’s development https://gist.github.com/nishio/270a066043e6a924062b045b44ba46e7

• ringo A lot of specific instructions needed to get people to write code that is more than just copying and pasting Tutorial code that falls around.

nishio I’m surprised that you could respond so well to my request to play a nice melody without using mp3s…scary…w nishio I gave instructions in English and got a reply in Japanese because of the behavior of “ChatGPT for middle and high school students”. I’m trying to use it myself for testing, and it’s because the system prompts include instructions to “answer in Japanese”. nishio Note that I have no experience with Astro or SolidJS, but even I can develop with GPT4.

[Yajiuma PC Watch Samsung has three confidentiality leaks due to internal use of ChatGPT - PC Watch].

According to reports, the use of ChatGPT was permitted in Samsung’s DS (Device Solution) division, which is responsible for semiconductor and other businesses, on March 11. According to the report, at least three incidents occurred in the 20 days since then. Two involved the leakage of equipment information and one involved the leakage of meeting content. The company is investigating the circumstances of the incidents and, as an emergency measure, has limited the upload capacity to 1,024B (bytes) per question to ChatGPT. If similar incidents occur in the future, the company may shut down the connection.

• It’s a delicate measure that may or may not be preventing confidentiality from being leaked while reducing productivity in a wide range of areas 🤔Limit upload capacity per question to ChatGPT to 1,024B as an emergency measure.

@shunkeen: As I was reading this, it occurred to me that if GitHub itself forks OSS that are used by a lot of people but no longer have people to maintain them, many people will be saved. I was thinking that if GitHub itself forks the OSS that quite a lot of people are using but there is no one to maintain them, and GitHub’s Tsuyotsuyo AI can automatically maintain them, there will be a lot of people saved.

nishio (CM here) The code I had GPT4 write yesterday and was surprised at how easily it worked when it was generated with test code. It was a code that hits kintone’s API. In other words, kintone is not a “lumpy black box” but an “open extensible design” and kintone users can benefit from the productivity gains of AI. nishio I think the fact that the documentation team has worked hard on the English version of the documentation is very much contributing to the fact that it is understood by GPT4. Because the code generated by GPT4 was using the foreign brand http://kintone.com

---

This page is auto-translated from /nishio/機密性と生産性向上の天秤 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.