Engineer
Major
That's nice and all, but the question is how to access it along with other Chinese LLMs for someone living in the West and depends on sevices in English.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
While big AI companies are trying to create huge models consuming billions in GPUs benefiting Nvidia pockets DeepSeek has really made use of innovation to create these models for cheap using millions instead of billions, so cheap in fact that they can open source the model AND STILL make money, is incredible. If more efficient AI models become the norm Nvidia could be in deep trouble.
OpenRouter offers access to the popular DeepSeek and Qwen Chinese LLM at competitive prices
Last edited:
Engineer
Major
I should clarify that the question is about accessing Chinese LLMs directly on Chinese servers without going through middlemen in the West.
I think it's important to get a practical definition of AGI first: a highly autonomous system that outperforms humans on most economically valuable (ie. useful) tasks. What a lot of techbros and laypeople are thinking fall more into the category of ASI.
With that in mind, several firms (incl. chinese) have actually built tools internally that scale, not quite infinitely, but close enough and they seem to have reached above human performance on most tasks - unfortunately at the cost of being much more expensive than hiring a few thousand humans to do it.
I talked to some folks who did work around this last year and the consensus seems to be there is no limit to how smart you could get a swarm of agents using different base models at the bottom end, if you don't mind burning through GPU cycles . At the time even this was a completely open question. It's still the case that no one has build an interactive system that _really_ scales - even the startups and off the record conversations I've had with people in these companies say that they are still using Python across a single data center.
IMO AGI the way it is defined above is now no longer a dream but a question of if we want to:
1). Start building nuclear power plants like it's 1950
2). Wait and hope that Moore's law keeps applying to GPUs until the cost of something like o3 drops to something affordable
Broader AI community haven't fully understood potential of agent swarms but top firms are deep into it already, and yes, some chinese researchers are well aware of it. Future breakthroughs such as getting AI compute to consume 1% of the energy it does today would be extremely significant.
AI community expects 2025 to be the year of the AI agents
In other news, Xiaomi has hired (poached?..) one of DeepSeek key founders
CNBC people gets it finally. Once you open source and commoditize leading edge AI, these tech companies who built their valuation on frontier tech are really not worth much.
this is a very good point I saw on X. It is not an exaggeration to say a lot of companies and funds have gone “all in” on generative AI. Just screwing some of them over can have a ripple effect on the US financial industry and overall economy
The woman interviewed by cnbc:
Deirdre Bosa is the anchor of “TechCheck,” CNBC’s tech-industry franchise. Prior to this, she was anchor of “TechCheck” in its daily program format that ran from 2021 to 2023. Previously, Bosa was a technology correspondent and contributor reporting out of Vancouver, Canada, for the network. She joined CNBC in 2012, covering the markets and economies of London and Singapore and co-anchoring “Squawk Box Asia,” “Squawk Box Europe,” and “Worldwide Exchange.” Earlier, Bosa was an anchor and reporter for CCTV News International based in Beijing and a contributor to Fox Business News. She began her career working for multinational corporations Barrick Gold in Toronto and Rio Tinto in Shanghai.