Anthropic employee posts Claude demo video but its a copyrighted YouTube video | Hacker News Reader

Anthropic employee posts Claude demo video but its a copyrighted YouTube video

twitter.com

24 pointslocalfirst12 comments

The title is incorrect because the claude demo video does not contain a "copyrighted youtube video". It contains a webgl graphics demo with very similar fractal graphics as the author of that tweet made a year ago [0]. In the claude demo the graphics react to mouse position, while the original does not take any input.

So my understanding is that at least one of these is likely to be true:

A. The code was used in Claude's training

B. Claude used browsing and came across the code

C. The Anthropic employee who posted that video gave Claude at least a part of that code earlier in the conversation

0: https://www.shadertoy.com/view/mtyGWy

Title implies the demo video was faked. It looks like Claude created code similar with a result similar/identical to one used in a shader tutorial.

Is there a more charitable way to read this than "Hamish Kerr is a shameless liar" or... ?

Yes. A completely reasonable read is: Claude was asked for something and responded with result based on work which exists online. So basically... Claude works at expected? (Same as other models)

The shadertoy example is just a few lines, so easy to train from. (And even that uses palette creation described on another blog)

I wonder how a disassembled and then scrambled code rates in this regard. Is it a piracy or a retell of a story? Is it something that perfectly makes sense? Because if that’s okay for some algo to retell copyrighted stuff without being in offense, then that’s a very bright future ahead of all.

No wonder companies stopped distributing software, and moved to the cloud. That way you can’t disassemble them right so?

Grab some popcorn and wait a bit. If Google-vs-Oracle can happen about reimplementing the API, we're definitely going to see a number of cases about the AI code in the future. And it's going to be way more interesting, because the threshold between copying and being inspired by is very fuzzy.

Not a good look for Anthropic. I mean everyone knows they are using stolen copyrighted data (same as Open AI and everyone else) but to be blatantly caught doing it is not really nice.

So was anything they did illegal?

I don’t get the problem in this case. If the guy wanted the code to be secret don’t publish it? Is there some confusion around how LLMs work here?

Is there some confusion about how copyright works?

In the case of AI, it seems to be a legal grey area that's actively er... evolving. ;)

Show HN: Every last letter of published code ever put to a readable and accessible medium is mine now. You can use any portion of it, for any purpose you like, just sign up to my Platinum Plan. Original authors shouldn’t have published it if they didn’t want me to resell it.

It's code from a tutorial. Mission accomplished, the LLM learned the lesson.

Reply on news.ycombinator.com