Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How to write stuff no one else can

thewritetoroam.com

Post reply on HN
Print view View original

21–30 of 143 posts

Re: How to write stuff no one else can

#21

Post by adtac »

hawski wrote: post #3
yamrzou wrote: post #2

This sentence caught my attention: > You can’t just Google these kinds of things. That’s why they’re so interesting. So what Google did is dilute the information advantage. The more interconnected we are, the harder it becomes to maintain that advantage.

That's also something I often encounter with ChatGPT. It often is very wrong about things I ask, because I often wonder and try things off the beaten path. That's our advantage against LLMs.

Purely probabilistically, trying things off the beaten path is just a matter of higher LLM temperature. Turning up GPT-4's temperature is basically an expensive /dev/urandom pipe today, but I don't see any fundamental reason why LLMs can't catch up. Maybe all it takes is tinkering with how temperature is calculated.

Re: How to write stuff no one else can

#22

Post by philipswood »

oldandtired wrote: post #15
wordpad25 wrote: post #11

Earlier quoted context omitted.

AI so much much more than retrieval since it knows how to correctly apply exist concepts to previously unexplored areas. You can ask ChatGPT how one might implement and enforce laws in a world where magic exists. Or time travel. Or whatever. Because it knows laws and it knows common fiction tropes and can reason well enough to surmise how to apply them together.

You make the following comment regarding artificial stupidity systems (AI) > since it knows how to correctly apply exist concepts to previously unexplored areas and yet these systems know nothing at all. Far too many people (including the developers of such systems) have failed to understand that none of these systems can go beyond the programming that humans have incorporated into them. Yes, they appear to exhibit c…

> and yet these systems know nothing at all. Far too many people (including the developers of such systems) have failed to understand that none of these systems can go beyond the programming that humans have incorporated into them.

> within the constraints of the programming involved in them

I think the point of machine learning is that programming isn't "incorporated into them".

These systems may not be intelligent yet, but we certainly didn't program them - the majority of their features and abilities arise from the training data they were exposed to.

Re: How to write stuff no one else can

#23

Post by simne »

And it is extremely hard to find person, who prefer the people who were deep in the trenches.

What I see most time, managers buy bright cover, very few bother to look deeper.

And it's easy to understand - as said wisdom, "No one fired due to buy IBM solution". When I agree, not many people are like IBM with near cloud-less clear reputation, but it's just simple and need less energy, to not try swim against flow.

Re: How to write stuff no one else can

#25

Post by aspenmayer »

wordpad25 wrote: post #12
wordpad25 wrote: post #11

Earlier quoted context omitted.

AI so much much more than retrieval since it knows how to correctly apply exist concepts to previously unexplored areas. You can ask ChatGPT how one might implement and enforce laws in a world where magic exists. Or time travel. Or whatever. Because it knows laws and it knows common fiction tropes and can reason well enough to surmise how to apply them together.

Wanted to give it a try. I didn't even consider many of these and doubt I would've come up with such a good list even if I had a lot of time. Much less than 1 second it took to generate the response. In a world where magic exists, laws and the legislative system would likely undergo significant adaptations to address the unique capabilities and risks presented by magic. Here are several ways in which these systems mi…

You basically just described the anime Witch Hunter Robin.

https://en.wikipedia.org/wiki/Witch_Hunter_Robin

Re: How to write stuff no one else can

#26

Post by snoman »

oldandtired wrote: post #15
wordpad25 wrote: post #11

Earlier quoted context omitted.

AI so much much more than retrieval since it knows how to correctly apply exist concepts to previously unexplored areas. You can ask ChatGPT how one might implement and enforce laws in a world where magic exists. Or time travel. Or whatever. Because it knows laws and it knows common fiction tropes and can reason well enough to surmise how to apply them together.

You make the following comment regarding artificial stupidity systems (AI) > since it knows how to correctly apply exist concepts to previously unexplored areas and yet these systems know nothing at all. Far too many people (including the developers of such systems) have failed to understand that none of these systems can go beyond the programming that humans have incorporated into them. Yes, they appear to exhibit c…

I get what you’re saying, and will echo that these systems are still just tools that enhance what we, as humans, are capable of doing. But I think you’re failing to consider that even comparatively simple systems can exhibit emergent behavior, well beyond their programming. You’re being unreasonably reductive and dismissive.

Re: How to write stuff no one else can

#27

Post by ethanwillis »

snoman wrote: post #26
oldandtired wrote: post #15

Earlier quoted context omitted.

You make the following comment regarding artificial stupidity systems (AI) > since it knows how to correctly apply exist concepts to previously unexplored areas and yet these systems know nothing at all. Far too many people (including the developers of such systems) have failed to understand that none of these systems can go beyond the programming that humans have incorporated into them. Yes, they appear to exhibit c…

I get what you’re saying, and will echo that these systems are still just tools that enhance what we, as humans, are capable of doing. But I think you’re failing to consider that even comparatively simple systems can exhibit emergent behavior, well beyond their programming. You’re being unreasonably reductive and dismissive.

The specific behavior you get is a result of the programming. what does "beyond the programming" mean? I think it's actually a meaningless statement.

I would love to see an actual example where a program does something not encoded in itself already.

Re: How to write stuff no one else can

#29

Post by chubot »

yamrzou wrote: post #2

This sentence caught my attention: > You can’t just Google these kinds of things. That’s why they’re so interesting. So what Google did is dilute the information advantage. The more interconnected we are, the harder it becomes to maintain that advantage.

If you want to maintain an information advantage, you should definitely not rely solely on the web. Simply reading a book is often 10x better. Comment from a few months ago:

https://news.ycombinator.com/item?id=38369583

One thing I've found comically underestimated is books. It's not whether the information is physical or electronic, but what actual corpus of information is available

...

So yeah I'd say 3 main repos of knowledge are: the open Internet, printed books, and Scihub, and many people today only use the first one

Though I'd also put "talking to the right people" and by extension "working at the right place" pretty high up there in getting novel information.

I think for a certain generation, Google probably diluted their ability to access other knowledge bases, and act upon that info. And for the generation currently learning how to learn, LLMs may do the same thing.

Re: How to write stuff no one else can

#30

Post by ben_w »

wordpad25 wrote: post #11
ssivark wrote: post #5

Earlier quoted context omitted.

Google, ChatGPT, Gemini, Perplexity, etc are all retrieval services (with some rudimentary remixing capabilities thrown in, for the latter few). Some human has to actually have interesting insights and write them. Retrieval services help with distribution & discovery. The less we’re connected, the harder it is to find the right audience who value your (niche) insights. Of course attribution and linking is important.

AI so much much more than retrieval since it knows how to correctly apply exist concepts to previously unexplored areas. You can ask ChatGPT how one might implement and enforce laws in a world where magic exists. Or time travel. Or whatever. Because it knows laws and it knows common fiction tropes and can reason well enough to surmise how to apply them together.

To an extent, but it's only about 85-95% correct even in the good domains, which is fantastic for a few steps… and then it makes a fundamental error and half the time doesn't understand corrections given by domain experts that would get it back on track.

This limitation is the only reason why all of us all have jobs, so do plan for your economic future on the assumption the AI get past this flaw with no warning — how many steps it can do before falling over isn't something I can predict, despite regular use of even the OpenAI playground before ChatGPT came out.

Post reply on HN