Heavy Networking 663: OpenAI For Networking

Would you give an artificial intelligence responsibility to write your router configurations? You wouldn’t. Not yet. But we’re not as far from that as you might think. For example, I gave the OpenAI tool ChatGPT the following prompt: “Generate a Cisco router configuration for two routers using OSPFv2. The first router should be in OSPF area 0. The second router should be in OSPF area 1. Area 1 is a stub area. The OSPF neighborship should form securely using keys.” The result was far from perfect…but it wasn’t a bad place to start. The AI had learned enough to know what I was asking for. The config it generated was missing some syntax and some dependencies, plus it added a few things I didn’t ask for like “redistribute static” and “default-information originate”. But it’s not hard to see how, given the right training on a narrowly scoped data model, the AI could indeed generate robust, usable configs. Our skill as engineers would be in the prompting–knowing what to ask the AI for to generate the configuration. I don’t even think this is that far off, considering where ChatGPT as a tool is today. Our guests are Ryan Booth and Phil Gervasi, both repeat voices here on Heavy Networking. They are employed by networking vendors you’ve heard of, and have been digging into the state of AI and its usefulness to networkers. Show Links: A Thousand brains – Amazon Training language models to follow instructions with human feedback – Arvix.org How ChatGPT Actually Works – AssemblyAI @that1guy_15 – Ryan Booth on twitter @network_phil – Phil Gervasi on Twitter