Thoughts on LLMs
5 minute read.
After skipping the Crypto and NFT discussions and mostly restraining from offering an opinion, I did end up running down a rabbit hole on the current AI trends.
I’m getting close to two decades of coding experience and recently spent about three days worth of working time (entrepreneur days) building a company-internal chatbot that is able to create and manage Gitlab Issues, CI/CD pipeline failures, draft emails and remember preferences. I ended up with a superficial understanding on what is possible and an intuition on what it is useful for and what not.
I was baffled at how easy it was to build these things on top of the OpenAI APIs.
In the end, I can summarize for myself: AI assistants are extremely appealing because you can provide tools and they just know when to use them. They can take over mundane tasks that nobody enjoys much, like summarizing conversations into a ticket and assigning the right person and milestones. And they do it much cheaper and faster than humans could.
They are also able to source the right information at the right time, compared to a human who will in the interest of time not evaluate seemingly unrelated content. An AI model cannot avoid processing the related context fed to it by embeddings.
# Note on AI Startups
Many startups out there are on a very high level “exposing” AI models to users in interesting ways. Their main tasks is UX and marketing, little on the AI model innovation side. But the plethora of potential use cases make this interesting nevertheless… sometimes.
Huge value is created when one model does everything rather than having a model or separate AI product for everything.