Your sexual experiences with Llamas?

[nice90sguy]

Are there other writers here who've played with LLM prompt engineering? I've got pretty deep into it recently, and I'm sure some other people here have too. Meta's latest offering, Llama 3.3 70B, is as as powerful as anything I've used commercially, and it's free (and runs locally, so you don't need to watch your language or curb your fantasies). It can run with your ideas, and at the very least, it's an incredible "spitballing" partner for your creative ideas.
I'm really interested to hear anyone else's experience with LLM's for creative writing, chat prompts, etc.

 

[THBGato]

I thought this was going to be a thread researching a forbidden love story set in a Tibetan monastery.

 

[TheLobster]

I thought this was going to be a thread researching a forbidden love story set in a Tibetan monastery.

And I was surprised that @nice90sguy, of all people, would be the one to brazenly flout the “no bestiality” rule.

 

[AwkwardlySet]

Meta's latest offering, Llama 3.3 70B, is as as powerful as anything I've used commercially, and it's free (and runs locally, so you don't need to watch your language or curb your fantasies).

Locally? You downloaded it? How much space does it take?

 

[StillStunned]

The only thing I know for sure about llamas is that they don't swallow.

 

[nice90sguy]

Locally? You downloaded it? How much space does it take?

Well admittedly you need a pretty hefty machine to run it - it’s a 40Gb download and your machine should have around 64gb memory and a dedicated nvidia gpu. High end gamers machines fit the bill.

 

[lovecraft68]

No, I prefer to do my own writing.

 

[AwkwardlySet]

Well admittedly you need a pretty hefty machine to run it - it’s a 40Gb download and your machine should have around 64gb memory and a dedicated nvidia gpu. High end gamers machines fit the bill.

I have 32GB of RAM, but I was asking out of curiosity.
But 40GB of SSD space doesn't compute at all... Is that packed data or something? Because there is no way you can fit 70 billion parameters in 40 gigs of space... that's like half a byte per parameter. Either it requires unpacking into much more SDD space, or it unpacks dynamically in RAM and thus the ridiculous RAM requirement? I mean, there is no PC game I know of that takes more than 16/32 GB...

 

[ShinyCrazyDi]

If you're just seeking writing prompts this seems like risky software to use, you don't really need to install AI to find writing prompts, ideas & scenarios. I would want to do some research about it too, with such a large footprint I'd wonder about undisclosed spyware/malware?

 

[lovecraft68]

I thought this was going to be a thread researching a forbidden love story set in a Tibetan monastery.

Lascivious ladies loving Llamas sounds like a good working title.

That's alliteration in the mighty Marvel manner! Excelsior!

sorry, my geek is showing.

 

[SimonDoom]

Without hats, of course.

 

[lovecraft68]

I’d steer clear of anything offered by a far-leftist like Zuckerberg. If it were Elon's, though...

Truth.

 

[onehitwanda]

No. Everything I write comes from my own mind; influenced obviously by others. An AI cannot create the world I see inside me, and I doubt it ever would be able to.

I'm sure people use AI to great effect. But I enjoy the process of writing and polishing a story, and the feedback I hunt from my very small list of trusted providers is a treasure I am not prepared to diminish by mixing in the hallucinations of code.

People view AI as a panacea. I view it as Roko's Basillisk.

 

[AwkwardlySet]

I’d steer clear of anything offered by a far-leftist like Zuckerberg. If it were Elon's, though...

Both are rich assholes, more or less. But Meta's Llama is fully open source and free, while X's Grok is only partially free. Without going into their political affiliations, life philosophies, and their history in IT, Open Source rich asshole > Non Open Source rich asshole.

 

[SouthernCrossfire]

I'm with Wanda on the writing. And I've never really been a big fan of basilisks or Skynets either.

 

[AwkwardlySet]

No. Everything I write comes from my own mind; influenced obviously by others. An AI cannot create the world I see inside me, and I doubt it ever would be able to.

I'm sure people use AI to great effect. But I enjoy the process of writing and polishing a story, and the feedback I hunt from my very small list of trusted providers is a treasure I am not prepared to diminish by mixing in the hallucinations of code.

People view AI as a panacea. I view it as Roko's Basillisk.

I think that the greatest disadvantage of using modern AI abilities is that it will negatively influence the growth of humanity.
Sure, it might help with some technological advances, but at its core, it's meant to be used as a crutch to avoid work, and with it, the learning and personal growth that comes as a result of such work. It will push society into something lazier and more hedonistic, into a sort of decadence that will result in the decline of humankind, perhaps not technologically, but in most other ways, probably yes.
Maybe I am being too pessimistic but human nature doesn't give me many reasons to think positively.

 

[yowser]

Fact is richer than diction.

 

[nice90sguy]

I have 32GB of RAM, but I was asking out of curiosity.
But 40GB of SSD space doesn't compute at all... Is that packed data or something? Because there is no way you can fit 70 billion parameters in 40 gigs of space... that's like half a byte per parameter. Either it requires unpacking into much more SDD space, or it unpacks dynamically in RAM and thus the ridiculous RAM requirement? I mean, there is no PC game I know of that takes more than 16/32 GB...

Yes , it’s around 4bits per param - a four bit Quant. Highly quantised - LLMs can be squished quite a lot without much degradation in performance.

 

[sinfantasy]

@nice90sguy - While I see your point for ideation... Just to be clear Lit does not allow AI-generated content. Your ideas can come from anywhere but writing had to be your own.

 

[onehitwanda]

I think that the greatest disadvantage of using modern AI abilities is that it will negatively influence the growth of humanity.
Sure, it might help with some technological advances, but at its core, it's meant to be used as a crutch to avoid work, and with it, the learning and personal growth that comes as a result of such work. It will push society into something lazier and more hedonistic, into a sort of decadence that will result in the decline of humankind, perhaps not technologically, but in most other ways, probably yes.
Maybe I am being too pessimistic but human nature doesn't give me many reasons to think positively.

people don't appreciate that you need to do the work to get good at something. Same for something like piano, same for painting, same for maths.

 

[AwkwardlySet]

people don't appreciate that you need to do the work to get good at something. Same for something like piano, same for painting, same for maths.

When I am teaching something new, I often make my students work for answers. I push them to think and try to reach a conclusion, and they often protest that I am torturing them instead of giving them the answer
How can they truly process and appreciate something that was handed to them on a silver platter? Something they didn't have to work for nor go through all the analytical steps in order to reach the right answer themselves? Answer given in such a way becomes information rather than knowledge and it will go away quickly. Yeah, AI is going to be a problem in the long run.

 

[nice90sguy]

The literary level of chat gpt is very low and is not a source of inspiration -its writing style is crass and riddled with trite metaphors and at best is a pastiche. So forget using any of its output directly.
But I referred in my o.p. to “spitballing” which doesn’t really require much intelligence or creativity - I remember actually doing this as a kid with my father who was a writer. I was just a sounding board.

Another thing that I’ve found useful is character. Instructing the ai to adopt a certain character , and then tweaking its behaviour as you go along , is really illuminating — it helps me, as the “prompt engineer”, to really learn how to define character clearly.

 

[TheWritingGroup]

When I am teaching something new, I often make my students work for answers. I push them to think and try to reach a conclusion, and they often protest that I am torturing them instead of giving them the answer
How can they truly process and appreciate something that was handed to them on a silver platter? Something they didn't have to work for nor go through all the analytical steps in order to reach the right answer themselves? Answer given in such a way becomes information rather than knowledge and it will go away quickly. Yeah, AI is going to be a problem in the long run.

"This class isn't about knowing the answers. It's about finding the answers. You'll never know all the answers, so you need to know how to get more answers, even after you're done with school." Me teaching stuff.

-Rocco

 

[onehitwanda]

"Light a fire for a man and he'll be warm for an hour. Set a man on fire and he'll be warm for the rest of his life."

 

[onehitwanda]

Still, those wary of pens will cling to their quills.

Quills still function when the last copper mine has failed, and those who have used them before can use them again.

Imagine having a proofreader, editor, and beta reader available 24/7

My position remains that I will trust the imperfect work of a human over the "perfect" work of an AI any day of the week, and twice on Tuesdays.