withwi 61 Everyone wants others to believe something, like others said, you should be careful what you take in from any model at all.
Are you more concerned about open weight Chinese models than closed weight US models?
The only real concern is the economic pressure this puts on the Anthropic and OpenAI.
A uneducated guess on my part is that they release these opensource models to get as many people away from the closed source AI’s so the bubble pops sooner and Anthropic and OpenAI goes bankrupt so they can take over the market afterwards.
There is a reason that slimy CEO of Anthropic wants to ban open source models afterall.
But honestly good riddance and annoyingly they will most likely take the rest of the economy with them.
If OpenAI and/or Anthropic goes bankrupt, it wouldn’t mean China was the only player left on the market. Microsoft, Google, Meta, or Amazon would just buy the leftovers and integrate the technology into their own portfolio. 1 Like
That’s going to keep happening, over and over again.
The same do closed source models with the added issue of running constantly to serve millions of users.
Meanwhile I just spin the thing up when needed.
The total power usage running AI on the edge + training vs cloud models + training leans heavly in favour of edge usage.
vic 65 I genuinely believe that’s not what the Chinese are thinking. For sure they want to serve the Global South in addition to serving primarily themselves since America sanctions the Chinese from using OpenAI and Anthropic models.
The fact that so many American firms pick up and use the Chinese models is just a by-product. it’s also due to amazing progress the Chinese models achieved in so limited time.
1 Like
thro 66 With the open weight models already available i think AI is here to stay irrespective of whether Anthropic or OpenAI go broke.
That genie is not going back in the bottle.
1 Like
Hard to say what the future for open-weight models will be. It’s not like open-source software, where a few dedicated people can run a project by themselves.
I don’t think the small models are going away, but I don’t know if we will continue to see +1T size models getting released as open-weight.
The AI from Proton (Lumo) is a mix of Qwen and GLM. If there is a valid business model for western companies reselling Chinese AI models, that could be the end of free open-weight models. The moment they can license the models, I don’t think they are going to give them away for free anymore.
I don’t think they will stop releasing the small models you can run at home on a GPU, but what we are seeing now, where close-to-frontier-level models are getting released as open-weight, I doubt that will last forever.
thro 68 I think we’re at the point now where models are close to “good enough” (for the foreseeable future) and tool use is where the frontier lies at this point.
The model itself doesn’t need to retain everything within its own data-set if it can intelligently query information via RAG, use local memory, etc.
vic 69 Came across this news in the past day or two:
Reuters developed its in-house LLMs known as ‘Thomson-1’ for its legal, tax and compliance staff in cooperation with Imperial College London.
Turns out Thomson-1 is based on Alibaba’s Qwen. It shows open-weight LLMs are possible to be customised.