Why we started "Karla" and what we are working on.
Our starting idea was rather simple: What could you use LLMs for that would have an immediate benefit to users today?
LLMs are great at writing, so this led to the question: Who is writing a lot? One answer is: journalists. But on a second look, you see how tricky it is.
Several problems emerged, or better, it crystallized what we would need to work on:
LLMs don't know about current events.
LLMs are available as chats, but people want a better UX.
You need to know how to get the best out of LLMs.
Let's quickly go through what we mean. This will give you an idea of what we are working on at the moment.
First and arguably most problematic: LLMs are, by design, static. They don't integrate new information after being trained on a dataset. News on the other hand is, by nature, very dynamic. So this doesn't work well together. To get the LLM to write news, you need to feed in the relevant information in a smart way. Or evolve LLMs, or generally the systems around them, on the foundational level.
Second: LLMs are most prominently used in chat apps. But the best app format to write articles are text editors, not chat boxes. Writing an article with ChatGPT involves a lot of copying and pasting between the chat box and the text editor. Now if it's just about writing some text with AI, there are already some cool tools out there, like Jasper or Writesonic. However, they focus on marketing articles. News requires a different workflow, and arguably, the problem of integrating current information can be better solved with a specific UX for journalists.
So that is basically what we are working on. Specifically, we have launched a super basic proof-of-concept text editor that turns information or news feeds into articles. You can try it for free at getkarla.com. We are working on an MVP that will hopefully be ready in the next 2-3 weeks.
Looking forward to more updates in the coming weeks.
... in case you wonder: This article was written by myself, but yes, we did use an LLM to polish its wording :)