Item — HN Reader

After the duty to document, comes the duty to expose it :

- How do you format that doc ? which is particularly relevant if you'd like to roam around. the author's Berlin guide is a good example [1]

- How do you find exactly what you want ? Some say that there's no longer (or lesser) need to document because we have LLMs. I would argue the exact opposite : Given their ability to streamline the retrieval process (into a naturally "human" chat), every documentation's worth is actually magnified by LLMs.

[1] https://nicolasbouliane.com/projects/all-about-berlin

2 comments

> Some say that there's no longer (or lesser) need to document because we have LLMs.

What will you train the LLM on? It has to get the information from somewhere. This is a point I'm supposed to make next month in a panel about using AI to help immigrants. An LLM won't make phone calls.

In a sense, we may have less need to create "formal, structured documents." Even today, many companies archive a large part of their informal conversations (meetings, emails, Slack...) in addition to "memos" and other short docs. The LLMs may get their data from there. Imagining a future where even open everyday conversation is recorded is a privacy nightmare, so this may not always be possible in less digitized environments (immigrants assistance ? I don't know)

Also, proponents of AGI believe that advanced forms of AI will require less training to be "creative," or at least to "infer/reason" on "low-level data" in order to create "higher-level" knowledge.

> An LLM won't make phone calls.

That is an interesting idea that I’m a bit worried could lead to more spammy ‘content’. I’d be surprised if the combination of LLM chat technology, voice recognition, voice generation, and affordable SIP trunks wouldn’t lead to someone deciding to start a website staffed by bots interviewing subject matter experts and writing articles based on those conversations.

Isn’t Google doing that in the US for booking tables at restaurants already?

I'd like to see it do the sort of journalistic legwork my job involves. This is a moat I don't expect AI to cross soon.

> After the duty to document, comes the duty to expose it

And then to maintain it. Which is hard.