Live data from Hacker News

Karpathy’s Pelican

twitter.com

321–330 of 461 posts

Re: Karpathy’s Pelican

#321

How much are the hobbit houses described in the book? The ones here look exactly like the movie

This one will be so deeply engrained in the model that no data cutoff will let it reimagine it. Though I suppose our human minds are 'poisoned' with the same image.

Re: Karpathy’s Pelican

#322
> Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story.

> it's an example where we go from "no one would ever do this" to "sure, why not, it's ~free".

How is two hours worth of token generation free? With tech revolutions things get cheaper/faster/better/doable, but there’s still real world limits. The advent of railroads made it feasible for the average person to cross the country, but it still cost a lot of time and resources. People weren’t crossing the country every weekend for fun just because it was now doable.

Why do we treat LLMs as ~free when we are generating things that weren’t doable before but have to invest more money than the Apollo program to build AI data centers let alone account for the operating costs?

Re: Karpathy’s Pelican

#323

It seems pretty clear that Anthropic models have been specifically trained to be good at generating three.js (JavaScript 3-D Graphics) code, so given current state of AI code generation in general, I don't find three.js models/animations as indicative of anything other than the model's ability to write three.js code. When Fable was first released the day-1 demos of it on Twitter (presumably from people who were given…

Yeah it's also specifically trained to disprove Jacobian conjectures, and so on...

Yes, it quite literally is. That’s what RLVR does.

Re: Karpathy’s Pelican

#324

Earlier quoted context omitted.

It was open sourced as Apache Wave when Google shut it down. Years later Apache moved it to read only because of low community activity. The archived git repo on GitHub remains available to clone and revive as a fork. https://github.com/apache/incubator-retired-wave

This doesn't explain the lack of marketing. Apache isn't exactly known for pushing tech. Google never tried to improve on gmail.

Google Inbox was amazing and I still mourn it regularly. The features they backported into gmail do not capture its essence.

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

Re: Karpathy’s Pelican

#325

I can forgive the modeling being godawful jank (windows floating in the air, disconnected from the house). But I expected it to have a better understanding of the text. Instead, we have Bilbo's "disappearance" interpreted as him magically transporting or cloaking, and similarly for his reappearance.

But this only seems wrong to you because you're familiar with the prior context. When there's only a single paragraph to work from, and it's a drily humorous text, why not employ comic literalism and lean into the perplexity with which his neighbors viewed him?

Isn't this almost certainly in the training set?

Re: Karpathy’s Pelican

#326

> Opus went off for ~2 hours and wrote 5500 lines of code that (procedurally) rendered the story. > it's an example where we go from "no one would ever do this" to "sure, why not, it's ~free". How is two hours worth of token generation free? With tech revolutions things get cheaper/faster/better/doable, but there’s still real world limits. The advent of railroads made it feasible for the average person to cross the c…

He wrote “~free” and also that he set a $10USD limit on it. For the result, I think it is fair to say it is approximately free.

Re: Karpathy’s Pelican

#327

Earlier quoted context omitted.

Humans are drawing pelicans riding bicycles now. Just google it and you will find 5 or 10 of them in the first few results. Including a t-shirt design. So it's a pretty much pointless test now.

Yet no LLM can actually do it. It's quite surprising actually.

What does this even test for? Can I use LLMs to directly generate machine code to replace my compiler? Or maybe I can use LLMs as a bare metal OS / scheduler to replace my machine's operating system and scheduler? It makes zero sense to test for that.

Not only that this so-called "benchmark" isn't economically useful, but that it tests for the sake of testing; and for attention.

To end this obsession with generating SVG pelicans, Quiver AI's [0] model is actually designed to generate SVGs from prompts and has done so for years.

So there is no need to continue with this un-serious pseudoscientific "benchmark".

[0] https://quiver.ai/

Re: Karpathy’s Pelican

#328

Earlier quoted context omitted.

This is a bizarre claim to make in this case, AI tech is used on all platforms universally.

iOS users spend dramatically more on e-commerce. I worked in e-commerce. Maybe it’s changed in the last 5 years. But that’s where the ops sentiment comes from.

More proportionally or more in total? Just wondering what the cost/benefit calculation is like when deciding how much to invest in the iOS user experience

Re: Karpathy’s Pelican

#330
post #155

AI is now somewhere between "Money for Nothing" ( https://www.youtube.com/watch?v=wTP2RUD_cL0 ) and "Knick Knack" ( https://www.youtube.com/watch?v=9uhM_SUhdaw ) in capabilities.

I find it interesting that people see Money For Nothing as evidence of technical limitations of the era.

It’s not. That blocky appearance and limited lighting was deliberate comic aesthetic choice as a response to budget, not a result of technical limitations. CGI in 1985 was enormously more capable than that — take a look at the stained glass knight in Young Sherlock Holmes, or The Last Starfighter from 1984. Things certainly could have been rounded, sculpted etc.

Post reply on HN