https://discuss.systems/@ricci/116338496516034352

TBF that's also true for a shocking amount of human-coded software...

Go out of business Bought out

Claude CLI code leak*

Words have no meaning anymore, apparently. Thank you

There is never a more permanent fix than the temporary kludge, duct taped in just to get the release out the door. That code is now a permanent, load bearing fixture that can never change.

Claude Code is the name of the product.

Claude Code is the name of the product.

If we're successful, we'll have the budget to rewrite it

Claude Code = The model + the CLI harness

The CLI harness code is what leaked

The model weights nor anything about the agents have not leaked

When we meet this deadline, we’ll have free time t… you want what new feature?

I did not fully understand what he meant in the toot so I asked an explanation directly to the source

Surprised it described itself as a war crime and didn't try to justify that

You'd wonder where language models learned all the slop.

The goal is probably to go public, with being bought out being the fallback plan.

If AI was actually good, it would've already rewritten that into a lean machine, not a duct tape artwork.

It seems like at first the people behind these AI companies were deluded enough to believe that they were really on the verge of creating AGI. When they realized that wasn't the case they changed plans and now they instead are building a house of cards as fast as possible and hoping that they can go public before it all collapses.

"I learned it from you, Dad!"

https://youtube.com/watch?v=CLU8VH870_k

(PDFA PSA)

Grazie.

Surprised it described itself as a war crime and didn't try to justify that.

It is not sentient. It has no sense of self or guilt or any of that. It's just writing back the most likely next word given the question and whatever context it has. It seems to have the leaked code in context? Or it could be bullshitting.

Our product doesn't have to work as long as it's the filthy public that's left holding the bag after we cash out.

Thanks for getting this clarified

It's probably just plagiarizing an article that used the turn of phrase to refer to spaghetti code. Or, just as likely, copying directly an article about the code leak.

Right, I asked the source of those figures (I didn't give the source in context, it just did a web search) and I got this: https://read.engineerscodex.com/p/diving-into-claude-codes-source-code

I find the conundrum of the 'Claw Code' scenario amusing.

LLM companies have argued they should get to ignore all copyright, and now that one of their code leaked, suddenly they care greatly about copyright.

But fine, their argument is that LLM digest of copyright infringements are 'fair use', so now that's been turned against them, using an LLM to launder their copyrighted material in precisely the way they declared doesn't count. So either they let it ride or undermine their own argument...

Or, and in my view the most likely option, they will get away with ripping off the little guy AND claiming "but bruh mah copyright" because they have very well paid lawyers and most western countries but especially the US have a two tiered legal system designed to protect the rich and punish the poor.

"In software engineering, that's a war crime"

LLM companies have argued they should get to ignore all copyright, and now that one of their code leaked, suddenly they care greatly about copyright.

Anthropic itself has argued that digitizing and using the digitized copies to train models is fair use, so long as:

• They don't redistribute the physical copies they bought
• They don't allow an end user to retrieve the contents of any one specific work at the user interface (if you ask Claude to spit out the entire text of a copyrighted work used to train it, it is designed to resist copying too much out of a single work)

So they don't argue that copyright doesn't count, exactly. They argue that copyright doesn't prevent model training from ingesting an entire copyrighted work, as long as it's done with so many other copyrighted works that any given original isn't a huge contributor to the model or its outputs.

There's tension in their positions, but not so much that it would totally fall apart.