Hacker News Reader

Comment by xorcist | Hacker News Reader

Right. LLMs are text production machines.

You can also view them as data expansion tools. They expand an input to an arbitrary long text. That text contains roughly the same amount of information as was contained in the input to begin with.

This is not very useful unless your work is content production. More useful is when the output length is shorter than the input. Then it is a (lossy) data compression tool. These are, as we all know, very useful for images and sound but now we have them for text too.

View on HN

This assumes one-shot generation. I don't use it that way, and I don't think many people who find it useful do.

I use it as a colleague I'm drafting with. I'll ask how a sentence lands, tell it to make a paragraph more friendly or more dramatic, have it argue against a point I'm making. The piece comes together over many turns, and the information arriving in each turn is mine — it's not expanding one prompt, it's helping me converge on something.

The information-content framing also proves too much: a copy editor adds no information either. Neither does a translator, or a ghostwriter. What they add is form, and form is most of what makes writing good or bad.

Many people make the mistake to think LLMs are stochastic parrots, but they are more like pianos, you play the keys and they make music, if you don't bring no new spices to the soup no wonder it comes out bland.

Exactly. LLMs regress to the mean. If you want don't want average, you need to feed the context to push it away from the mean.

Me? I tell it what to write, establish the general picture, then we write together.

Obviously I make software, not prose(not yet anyway). I would say, I contribute 95% of the “original thought”.

I really hate the whole thing about companies promoting using LLMs to change one-liner emails into this long multi-paragraph thing that ends up saying almost nothing ... just wasting everyone's time in reading them (with an extra sprinkling of hallucinations).

I believe you are supposed to use an LLM on the receiving side to compress them again.

This is supposed to help our economic growth during the coming decade.

/s I hope

What part of it do you think was sarcastic? Economic growth is measured by the resources humanity wastes.

In webcomic form: https://marketoonist.com/2023/03/ai-written-ai-read.html

LLMs have created more billable hours for me as an attorney because the LLMs convince the client that they need to know more about irrelevant issues, how specific statutes may apply, and other such things. They may ask about specific procedural issues that they would otherwise have no idea about that I then need to explain in greater detail. It reliably bloats things that would be a .1 into .5 or more.

>They may ask about specific procedural issues that they would otherwise have no idea about that I then need to explain in greater detail.

My favorite part about LLMs is that their insufficiency as legal aids (Aides? Aids.) seems to be slowly cluing-in the general public as to how Byzantine and "inside baseball" our legal system has become. (This is a bad thing.)

It's a big part of why I'm strongly opposed to common-law systems (as opposed to civil law). All the case law, all of it becomes, well, law. And then you get the whole "legislating from the bench" critique, which, while valid critique, is only a big problem because it's a common-law system.

Yes, I realize many (most?) legal systems are actually hybrid, but they do tend to err pretty strongly into one direction or another.

I agree that the output is more likely to be valuable when it’s shorter than the input. And when they use tools to do research and try things, it can be a lot of input that comes from all sorts of places.

>Right. LLMs are text production machines.

I use them as text consumption machines. They read and watch a lot of the books and movies i have no time to. And from the TLDR I decide whether the full thing is worth watching.

That is what humans are for!! Ask people!

There is no way a summary of these things fron an LLM has the correct context.

(not really frustrated at you, as much as where society is going)

That is not how it works because of how bloated everything is. There is rarely a movie, book or series lately that are not 2 times longer than needed. So you are forced to go for aggressive recaps.

Most of the non pulp books published - contain at least two times as many pages as needed for the story or narrative.

The information density in the modern world is absurdly low.

Yes, I agree on modern trends for that. There are some modern authors not too bad in that respect, but it may be why I find older sci-fi books more appealing.

Reply on news.ycombinator.com