> Here's a new trend happening these days. Upon releasing new non-fiction books to the general public, authors are simultaneously offering an LLM-based chatbot box where you can ask the book any question.

Can you link to an example?

I saw at least two examples of this here on HN. One of the books was about tech entrepreneurship 101, and I remember asking how to launch if you're a sole developer with no legal entity behind the product. I remember the answer being fairly coherent and useful. I don't have the URL handy, I suspect if you search HN for "entrepreneur book" you'll find it.

Some of the responses I've had so far to this are remarkable. Kind of scary.

Fascinating, thanks

I couldn't get "designing data intensive applications" to explain to me how to design a graph database (from scratch, without using existing graph frameworks or technologies), but it only suggested reasons why graph databases are useful and the properties I have to keep in mind while designing it. I want to know how I can build one in practice.

Using a prompt like "Tell me how to build a graph database from scratch. Specifically, how to design the data model, implement the data storage layer, and design the query language." only gives a very vague answer. Sometimes it suggests using existing technologies.

Anyone know what I'm missing?

I don't really think that book is about building a graph database from scratch

You're probably right.

One of my initial prompts mentioned graph databases as an example of a scalable system, so I wanted to ask it about the design properties that make it so. I figured that because it was a book about designing systems, it could give me an outline of how a graph database works in practice.

It's pretty annoying how the site erases your prompt once you receive your output. By the time it finishes loading I've half forgotten what my original question was.

Incredible results to my questions. Do these work by finding similar pieces of text from a vector DB, and then embedding those similar pieces of text in the prompt? The answers I'm getting seem to be comprehensive, as if it has considered large amounts of book text, curious how this works as there's an OpenAI token limit. I've heard this is what tools like langchain can help with, so maybe I should play around with that as this all seems like a mystery to me.

For more context, hn post by maker here: https://news.ycombinator.com/item?id=34635338

Wow! I guess some of the answers on questions I tried were pretty generic, but I can already see a value in it, and it's only a beginning.

How legal is something like this?

Genuinely unknown at this time. At some point this will be litigated in court, and if the parties don't end up settling, we'll then have some precedent that can answer your question.

> There is no good reason this should not work everywhere else, in exactly the same way. Take for example a large retailer who has a large internal knowledge base. Train an LLM on that corpus, ask the knowledge base any question.

Since LLM’s can’t scope themselves to be strictly true or accurate, there are indeed good reasons, like liability for false claims and added traditional support burden from incorrect guidance.

Everybody is getting so far ahead of the horse with this stuff, but we’re just not there yet and don’t know for sure how far we’re going to get.

"LLM’s can’t scope themselves to be strictly true or accurate"

This isn't true though the techniques to do so are 1. Not as yet widespread 2. Decrease the generality of the model and its perceived effectiveness.

I'm interested to hear what these techniques are. Decreasing the generality will help, but I fail to see how that scopes the output. At best that mitigates the errors to an extent.

conformal prediction

Predicting a set of answers to some confidence interval would still result in hallucinated answers.

low probability answers get shot to a human and reviewed for model improvement.

Requiring the answers to automatically verifiable, or having answers be inputs to a reliable query system?

I’d rather use the reliable query system!

If they are accurate for ~80% of the questions, they will be as accurate as any 1st or 2nd line help desk.

Hmm. Hypothetically if a human on first line help desk gives advice that is so completely bad as to be a crime, are they liable or the company? Because I guess a chat-bot would definitely not be liable.

How often is your fast food order 100% correct?

Correctness isn't one-dimensional. A wrong fast-food order might substitute or leave something out. There's essentially no chance the employee will swap in a random product from some other store.

But in this example the AI could hallucinate a statement attributed to you it actually formed by putting together reddit comments.

Sometimes they accidentally include tomatoes but they rarely include bombs.

Wrong food could be equivalent to bombs in worst case for who have allergy

> Since LLM’s can’t scope themselves to be strictly true or accurate

Bing tries to solve this and succeeds somewhat. It will insert Wikipedia style citations against each of its claims. You can visit them and verify the statement if you want. And I do it often.

No reason why a future DocAI can't link to specific sections in internal documents whenever it answers a question.

Haha, you work at Algolia. RIP F o7

How did GPS tracking companies survive Google and Google Maps? I think there will probably be many niches to explore even as the big names work hard to compete and eventually commoditize LLMs

Like TomTom?

I haven't heard the name TomTom in over a decade.

They’re still around