Parsons

If you’re like me, sometimes you just want to have a conversation with your network. When things go wrong, and you’re searching for a workaround, it would be nice to just ask, “Do you know why my network is so slow?” If you join me in that frustration, there’s good news! And, if you thought the solution would have something to do with AI, well, you’re right! There’s a new AI technology called Generative Artificial Intelligence (GenAI) that uses Large Language Models (LLMs) trained on massive amounts of data. When augmented with information about your network, GenAI chatbots can be used to have a conversation with you about anything contained within the LLM. The most recognized chatbot is ChatGPT, by OpenAI, but there are many others, and more are being developed every day (e.g., (Claude by Anthrophic, Llama 2 by Meta, Bing Chat by Microsoft, and Bard (LaMDA) by Google).

Recently, companies that provide Network Performance Monitoring (NPM) tools have started offering tools that use ChatGPT, and the results are going to change the industry. The best part is that anyone can use ChatGPT today and see significant improvements in their NPM. We have only begun to scratch the surface, but ChatGPT can provide many amazing benefits that were previously only available by asking experts, and even then, we may not be able to provide answers as clear and accurate as ChatGPT. For example, here are a few common tasks that ChatGPT can help you with:

• Hardware-related issues – ChatGPT can provide suggestions as to possible causes of the issues and then provide step-by[1]step instructions on how to fix them. It sometimes provides configuration scripts and other instructions to guide you to a fast solution.

• Network issues – ChatGPT can be your diagnosis assistant. It will offer recommendations, provide alternative approaches, and allow you to ask questions along the way if you don’t know how to carry out its instructions.

In the very near future, network administrators will not be able to compete without this technology. This brings up the first question everyone asks, “will it take my job?” Not anytime soon, but anyone who does not embrace it will definitely be left behind. Network maintenance and security are too critical to be turned over to an AI without human oversight. We must ensure that anything we do that touches our critical infrastructure has strong human controls with the authority to push the “off-button” when necessary.

"Recently, companies who provide Network Performance Monitoring (NPM) tools have started offering tools that use ChatGPT and the results are going to change the industry"

How do you troubleshoot a network issue that involves multiple layers of the OSI model

As an AI language model, I don't have practical experience in troubleshooting network issues. However, here are some general steps that can be followed to troubleshoot a network issue that involves multiple layers of the OSI model:

1. Identify the symptoms: The first step is to identify the symptoms of the issue. This could include slow network speeds, connectivity issues, or other problems.

2. Determine the affected layers: Once you have identified the symptoms, determine which layers of the OSI model are affected by the issue.

3. Gather information: Collect as much information as possible about the problem, including error messages and logs.

4. Analyze network traffic: Use a network analyzer tool to capture and analyze network traffic to identify any anomalies or errors.

5. Check hardware and software configurations: Verify that all hardware and software configurations are correct and up-to date

6. Test connectivity: Test connectivity between devices at different layers of the OSI model to isolate where exactly is causing an issue.

7. Check firewall settings: Verify that firewall settings are not blocking any necessary traffic.

8. Consult documentation and support resources: Consult vendor documentation or support resources for additional troubleshooting steps specific to your equipment or software.

9. Implement solutions: Based on your analysis, implement solutions such as updating firmware or reconfiguring settings to resolve the issue. 10. Monitor performance: After implementing solutions, monitor performance over time to ensure that the problem has been resolved permanently

To demonstrate the power of ChatGPT, I asked the following question without any prompt engineering (I’ll explain that later), and it gave me this answer:

How do I determine the affected layers? To determine the affected layers of the OSI model, you can follow these steps:

1. Identify the symptoms: First, identify the symptoms of the network issue. This could include slow network speeds, connectivity issues, or other problems.

2. Analyze error messages: Check for any error messages that may indicate which layer is affected. For example, if you receive an error message related to TCP/IP protocol, it indicates that there is an issue with Layer 4 (Transport Layer) or Layer 3 (Netwo rk Layer).

3. Use a network analyzer tool: Use a network analyzer tool to capture and analyze network traffic to identify any anomalies or errors at different layers of the OSI model.

4. Test connectivity: Test connectivity between devices at different layers of the OSI model to isolate where exactly is causing an issue.

5. Consult documentation and support resources: Consult vendor documentation or support resources for additional troubleshooting steps specific to your equipment or software. By following these steps, you can determine which layer(s) of the OSI model are affected by the network issue and focus your troubleshooting efforts accordingly.

It is pretty amazing that it provided this answer in a few seconds, and it includes a comprehensive overview of all of the steps related to my question. This initial answer doesn’t provide a lot of detail, but it helps a technician see the big picture immediately and possibly identify other areas to investigate. Now, let’s say you aren’t familiar with how to perform Step 2. Let’s ask a follow-up question:

The great feature of using ChatGPT is you can have a conversation, and it remembers what you are discussing. It continues to understand and answer within the context of the conversation, unlike using a search engine that only completes one search/answer at a time.

I previously told you that I would discuss “prompt engineering,” so what is it? Let’s ask ChatGPT:

What is AI language model prompt engineering?

AI language model prompt engineering refers to the process of designing and refining prompts for an AI language model to improve its performance. In the context of AI language text by the model. Prompt engineering involves selecting and constructing prompts that are most likely to elicit high - quality responses from the model.

The goal of prompt engineering is to optimize the performance of an AI language model for a specific task or domain. This involves experimenting with different types of prompts, adjusting their length and complexity, and fine - tuning other parameters such as temperature or top - k sampling.

Prompt engineering is particularly important for applications such as chatbots, question answering systems, and natural language processing tasks where generating accurate and relevant responses is critical. By carefully designing prompts that provide relevant context and guidance for the model, developers can improve its accuracy and effectiveness in these applications.

In a nutshell, prompt engineering is the process of adding contextual information to your conversation to reduce the scope, yielding more specific answers to your questions. When you provide information like the hardware, software, versions, etc., before asking your questions, ChatGPT’s answers will be tailored to your needs. Many companies (like Parsons, who I work for) have established internal capabilities to assist employees in carrying out very lengthy prompt engineering, which is called Retrieval[1]Augmented Generation (RAG). In RAG systems, users can upload massive amounts of data, documentation, user manuals, etc., and then ask questions about the material. This is very powerful and can save countless hours previously wasted searching for answers. For NPM, your network status can be used in a RAG configuration to ask questions about the real-time status of the network. This brings up a new issue: information security. Each time you enter your prompts, data, and questions into ChatGPT, it is stored on their servers and has the potential to be used in future model training sessions. To avoid this, we have established separate instances of the environment so all of our information is secure and controlled.

If you take a few minutes to sit down with ChatGPT and ask some questions, the power of the technology will quickly become evident. If you’d like to give it a try, visit chat.openai.com and start your generative AI journey.

Remember, though, ChatGPT is open source, so protect your proprietary work. Then, go one step further and implement an isolated model for your business to support collective use and data input. Once you have found the right ChatGPT use cases to support your business, you’ll likely find that the pace of productivity moves in leaps rather than steps – and it’s free (for now). Have fun, be smart, and get chatting!