ml work

With the current generations of LLMs, something that I see changing soon is the way we manage local information like a company's knowledge base. At the moment, documentation as an act has a lot of accidental complexities. Few examples:

  1. You have to write information to make it understandable for the reader. A lot of poorly written documents fail in their purpose when they encounter a reader in need.
  2. You have to maintain a good documentation tree (or graph, whatever your jam is) to make information accessible. Or if you are coming from the other side, your tool needs to have really solid search.
  3. Connected with previous point, you have to ensure that addition of new information is consistent with past knowledge. In case of overrides, you have to ensure the final state is correct.

A few of these are classically considered an important role of writer and the whole discipline of writing, but I believe it will be foolish to be a luddite and not accept these to be accidental complexities of writing, specially in case of pragmatic knowledge bases. The only essential complexity here is to input information in a minimal viable form and get it out consistently and correctly as needed. Even this could evolve to a stage where documentation systems are not about fancy information indexing but co-creation of knowledge.

If I were in a company like Notion, I would probably try pushing towards a race here. But I am content with the fact that people are already picking up a few interesting ideas that could lead to the next wave of documentation systems within 1-2 years. Probably we will see systems starting with semantic search and generation in isolation, then moving more and more towards managing the full documentation life cycle.

At some point in time, I believe the idea of local knowledge will disappear. But till then I am betting there will be a set of product waves with the common pattern of general intelligence applied over local knowledge, one of which will take documentation to its proverbial grave.

The next important question is: Are we also approaching the death of writing?