Its also a weird way to cede control of the market to the foreign model vendors, because I'm reasonably sure that DeepSeek et al aren't subsidising tokens to the same extent that the big 3's subscription models have been.

▲

xienze 2 days ago | parent [-]

I think there's "sort of" a moat for non-Chinese vendors. As much as people distrust the US right now, I think deep down inside everyone knows that the second you let a Chinese provider do inference on your codebase they're gonna suck up every bit of it. But hey, cheap tokens, right?

So you'll probably never see government customers allow that and neither will a lot of commercial customers.

▲

2ndorderthought 2 days ago | parent | next [-]

Why do we assume us providers aren't doing the same? Also all the Chinese providers are giving open weight models. Many you can run locally.

I don't see the risk. If your code is easily AI generated you don't have a moat anyways. A Chinese competitor probably won't have as easy of a time as a US one of you operate in the US

▲

Laughsalot 2 days ago | parent [-]

The US has robust IP and trademark law that allows companies some amount of chance to find a legal remedy to anyone who clones their business. China is notorious for protecting local companies from foreign IP suits.

Further, at a lot of companies, the risk has to be acceptable to shareholders and auditors. Perceived risk is often a more powerful motivator than actual risk.

	▲	isoprophlex 2 days ago \| parent \| next [-]
		> robust IP and trademark law lmao tell that to the artists, authors and foss contributors whose work has been cloned into the llm oracle
	▲	dijit 2 days ago \| parent \| prev \| next [-]
		I don't think that's as true as you think it is. I mean, "Copyright Infringement" famously does not translate to Mandarin; but we have Amazon ripping off best sellers in their marketplace pretty brazenly and Apple "sherlocking" applications -- that's even where the term comes from. The models themselves are trained on a corpus of material that was obtained with dubious legality... though I suppose some argument could be made that they're forced to bend the models because of that. I'd be more wary of these models terms and conditions granting a license to themselves for everything they come into contact with: nobody is reading these licenses it seems. Copilots old one only allowed for "being inspired" by the output, despite they themselves producing an IDE of some kind which allowed you to make complete projects from suggestions: directly breaking their own T&C's.
	▲	2ndorderthought 2 days ago \| parent \| prev [-]
		Once your code, images, etc pass into the slop machine it is owned by whoever generated it later. Obviously they would need a new logo, llc, and some ui theme tweaks. Otherwise none of these AI coder products would exist. Also, how long do you think openai, Microsoft, Google, anthropic, etc could delay a lawsuit while you pay hundreds of thousands in legal retainers? 5 years? 10?

▲

swiftcoder 2 days ago | parent | prev | next [-]

> As much as people distrust the US right now

From the perspective of someone currently living in the EU... I'd say thats pretty much a wash (or even slightly tilted in China's favour) for folks outside the US

▲

dannyw 2 days ago | parent | prev [-]

Fortunately there’s plenty of open weight models that are just safetensors and you have a wide variety of providers to choose from, as well as just hosting it yourself.