After a long hiatus I am finally back. There is a story I can’t wait to tell. For now, lets talk about large language models (LLMs)
I have given some thought to LLMs specifically on the source of data LLMs consume. For some context I heavily use chatGPT and GitHub copilot during my hobbyist development. I am impressed. I’ve thought about how these LLMs are trained. They are given access to content created largely through text, such as open source program source code available to anyone. I wondered what sort of impact LLMs tooling may have on the type of writing people do.
I concluded that documentation for programming languages will be pressured to cater towards being “understandable” to LLMs. This is kind of cool because it is often joked in software engineering that “no one reads the documentation” — but a LLM will! The exciting thing about LLMs is the ability to ask questions of data.
A minor aside: I believe that it is inevitable that websites develop their own LLM chatbot so that a user can use natural language to “ask data questions” for specific business domains. For example, an e-commerce website that creates a marketplace for service providers to do jobs for clients would be interested in hearing about your roofing project and routing you to the right collection of service providers. The value proposition is being able to let the LLM parse different aspects of a job on behalf of the company.
I think that general human knowledge being accessible through LLM clients is ultimately a good thing. I would be sad if this technology is withheld from the general public for monopolistic means.
I wish the LLMs that I use didn’t have their knowledge date frozen.
Let me know what your thoughts are on LLM tooling in the comments below. I’ll have more writing about generative a.i. soon!