この記事の日本語版はこちら！
0. Who am I?
Hello!
I am Bahador Bakhshi. I am a member of NTT DOCOMO’s 3GPP SA2 Standardization team. 3GPP is the organization that standardizes mobile networks, e.g., 5G. 3GPP is composed of working groups and SA2 is the group that defines the architecture of the network. Companies are sending their delegates to the 3GPP meeting to contribute standardization.
Nowadays everyone is discussing about AI. In this post, I am going to explain how AI can and cannot help people like me. If you are a delegate in a standardization organization, I believe you will find this article useful for your job. Even if you are not a delegate, the idea of creating customized AI chatbot for yourself would be interesting.
//For our 6G initiatives around AI, please also check DOCOMO Euro-Lab's YouTube video!
- 0. Who am I?
- 1. What does a Standardization Researcher and 3GPP Delegate do?
- 2. What Challenges Do Delegates Face?
- 3. How off the shelf LLMs (e.g., ChatGPT) Can Help
- 4. But Why the General Off-The-Shelf Chatbots Are Not the Perfect Solution
- 5. How to Customize ChatGPT to be More Suitable for Standardization Delegates
- 6. Should I Worry About Losing My Job?
- 7. Final Remarks
1. What does a Standardization Researcher and 3GPP Delegate do?
OK! Let’s start with my job in NTT DOCOMO.
Behind every network generation, there are standardization delegates, like me, who work together across companies. 3GPP delegates prepare and submit technical contributions, study new ideas, and review documents from other companies. We try to understand different viewpoints and much of the work is about finding a solution that different companies can accept. To be effective, a delegate needs strong knowledge of system architecture, and how mobile networks work.
// You can find some of the past articles here to find how it actually look like!
2. What Challenges Do Delegates Face?
Delegates typically face challenges in their work. We need a large amount of background knowledge as specifications are linked together. To understand a proposal, a delegate may need to jump between several specifications to see what the problem is and how to resolve it. Every meeting brings a flood of documents that we must read, summarize, and compare. We also need to track how ideas move through from study phase to actual standards, which can be difficult in large legacy content that has changed many times. For newcomers, the amount of information and the speed of work make the learning curve very steep and difficult.
3. How off the shelf LLMs (e.g., ChatGPT) Can Help
Off-the-shelf LLMs like ChatGPT can support SA2 delegates. They can explain background ideas and give an overview of how different parts of the 5G/6G system work together, which helps us understand a topic before we jump into the detailed specifications. LLMs can also summarize long technical documents, so delegates can quickly see the main point without reading every line. Delegates can also use LLMs to brainstorm alternative solutions, for example asking how a proposal could fit better with principles.
4. But Why the General Off-The-Shelf Chatbots Are Not the Perfect Solution
Tools like ChatGPT are not perfect for delegates. Their answers are not always fully reliable because they are trained on large amounts of internet text, not only on 3GPP standards. As generative models, they try to produce an answer even when the topic is unclear or not fully standardized, which can lead to mistakes or statements that do not match the real specification. LLMs are predictive models, not verification systems, so they cannot judge what is true or false. Delegates who need exact and standard-aligned information, this means that general chatbots must be used carefully.
5. How to Customize ChatGPT to be More Suitable for Standardization Delegates
To overcome the limits of general chatbots, ChatGPT offers several features that allow SA2 delegates to customize the tool and make it much more useful for standardization work. The first option is to create a custom GPT, where the delegate can upload a standard specification as the knowledge base. This helps the model focus on the right material and stay closer to 3GPP concepts. The second option is to use Projects, which work like organized folders containing documents including the standard specifications. The delegate can then ask questions directly about these files, and the model will ground its answers in that context. In the following, I am explaining the first option. The second one is easy and you can try by yourself.
OK, let’s create our own GPT for 3GPP Specifications!
In the ChatGPT user interface, in the left-hand side menu, there is “GPTs” and “Explore”. Clicking on this will let us find some customized GPTs; indeed, some other people already created GPTs for 3GPP, but we want to make one for ourselves. On the right-top corner there is the “Create” button!
Now, we can create a customized GPT. The ChatGPT is going to ask some questions and based on the responses it will configure the GPT.
The first question is about the main purpose of the GPT. You see my response to this question.
Then, it asks about the name of this GPT.
The next question is quite important since it defines the overall behavior of the GPT. In this example, I emphasized accuracy which is crucial for standardization.
Then, there is question that somehow limiting or restricting the GPT. For standardization, I think the most important thing is to make sure that the GPT is not going to generate something which are not defined in the specifications. So, I provided the following answer.
The last question is about how the GPT is going to generate output. Here, I prefer to give very direct and concise responses.
Now, we almost created the GPT but one (and indeed the most important one) is in the “Configure” tab.
In the “Configure” tab, the “Name”, “Description” and “Instructions” are already provided based on the responses that we provided in the “Create” tab. If we want, we can add more instructions to ensure that the GPT should not go beyond the text that will be provided in the knowledge base.
The next step is to upload the 3GPP specification as the knowledge base of this GPT. In this example, I uploaded the recent version of the three main SA2 specifications TS 23.501, TS 23.502 and TS 23.503.
You can also enable/disable some capabilities. I disabled all since I don’t expect that this GPT search web or generate image.
After these configurations, we should click on the “Create” on the top-right corner. That shows the window that allows us to decide how to share the GPT. In this example, I keep it only for myself and then “Update”.
Now, after the “Update” the GPT is ready.
This is the first example of the customized GPT!
As you see, the GPT is ready; and I can use it to check and verify the concepts defined in the uploaded specs. I am not going to provide detailed technical examples of SA2 specifications in this GPT since most probably you are not a SA2 delegate!
It should be noted that there are some limits (based on the subsection) on the number of files that can be uploaded into the knowledge base. So, do not expect that you can upload whole 3GPP specifications in the knowledge base to create super knowledgeable delegate 😉
6. Should I Worry About Losing My Job?
Some people worry that AI tools like ChatGPT might replace their job, but I think the core parts of my job cannot be automated easily now. Standardization is not only about reading documents; it is about understanding complex technical trade-offs, making judgments about architecture, and knowing how a change affects the whole system. Delegates must discuss proposals with experts from many companies, negotiate compromises, build consensus, and sometimes defend their company’s strategy in meetings. They also need creativity when designing new features, political awareness when different interests collide, and communication skills to explain ideas clearly and respectfully. These human abilities are at the heart of 3GPP work, and current AI cannot replace them. The currently available AI can support the process, but it cannot lead it.
So, no worries 😊 I will be able to continue standardization of 6G. Let’s see how intelligent AI will be in future; maybe, they can replace delegates in standardization of 7G 😉
7. Final Remarks
The work of a 3GPP SA2 delegate is complex, demanding. Delegates must deal with large amounts of information, understand many specifications, and work with experts. AI tools like ChatGPT can make this work easier by helping with reading, summarizing, organizing information, and preparing ideas. For the technical aspects, the general off-the-shelf LLMs are not suitable tools but ChatGPT provides facilities, including customized GPTs that we can create a new GPT which is focused on the 3GPP standards that we provide as the knowledge base.
As 3GPP moves toward 6G, the amount of work will grow, and AI will become an even more important. In the end, the future of standardization will be shaped by people who know how to combine their expertise with the power of AI. These “AI-augmented delegates” will be able to work faster, understand more, and contribute even more effectively to the next generation of mobile network standards.
If you also have any experience customizing AI tools, e.g., via prompt engineering, for your work/life, please share with us in the comments.
