I mean I'm producing more and better art and software each month than I used to do in a year. I feel like I'm flourishing because of AI. Tell me I'm wrong?

▲

k12sosse 3 hours ago | parent | next [-]

But does it suck?

▲

jaybeavers 4 hours ago | parent | prev [-]

You’re not wrong. Me too. I spend most of my waking hours using AI for new product development.

But can you bring it to market and turn it into a reliable revenue stream? And will this still be true in a few years when the duo or trio -opolies corner the AI market enough that the $200 subscriptions go away to be replaced by API pricing only?

I’m already seeing my subscription use slow to a crawl during peak hours so I now code before 8 am or after 6 pm. And turning on ‘Opus Fast’ with API pricing is already financially impossible for me as a small business.

Ironically, China may be the savior here, with open source models that may be good enough to ‘put the means of production in the hands of the working class’.

I mean, is QWEN a 4d chess game to replace capitalism with communism? Who knew (outside of the CCP central committee)?

▲

hollerith 4 hours ago | parent [-]

>is QWEN a 4d chess game to replace capitalism with communism?

No. The Chinese labs pay much less to train their models because they distill Western models. Western labs can't do it that way because they'd immediately be sued.

To distill a model requires high-quality prompts (and responses). Here's how the Chinese labs get these prompts: when a user sends a prompt to www.kimi.ai, Kimi immediately forwards it to Claude or another leading US model (using a vast network of laundered subscriptions), waits for the response from Claude, then forwards that response to the user via www.kimi.ai. All the Chinese labs do this. Of course, they never inform the user that their conversation is being routed to a US model.

Beijing is angry at them because they did this indiscriminately without filtering out the conversations of high Chinese government officials and military officers, so now the US labs have those conversations, which contain information useful to US intelligence because the officials and officers assumed the conversations would remain inside China's borders and that the Chinese labs would care about confidentiality.

The Chinese labs release model weights under permissive licenses to get any attention and usage share at all for the model.

▲

jacquesm 3 hours ago | parent [-]

> The Chinese labs pay much less to train their models because they distill Western models.

That's nonsense.

	▲	hollerith an hour ago \| parent [-]
		I'm going to take your comment as a request for supporting details. I believe what I wrote is an accurate summary of information in a report published this month by Anthropic. I haven't seen the report, but have seen this next summary written by someone who I trust to summarize accurately: https://thezvi.substack.com/p/the-bad-guy-with-an-ai-named-c... OpenAI reported similar behavior last year. The assertion that the other (two) leading US models are being distilled in a similar manner is an assumption on my part. Where I wrote, "Claude or another leading US model" I initially had just "Claude", but it felt silly to imply that the same thing is not happening to the other leading US models.