Don't conflate the model and the API offered by the company itself (in China) - obviously they've got to comply with Chinese laws but the model is open source, you don't need to use their API when it can be self-hosted or hosted elsewhere without giving them any of your data or indeed being subjected to whatever additional censorship they may be obliged to add to their API. People absolutely can use it (and indeed are doing so).
Chatgpt - Seriously good potential (or just some Internet fun)
- Thread starter Begbie
- Start date
Don't conflate the model and the API offered by the company itself (in China) - obviously they've got to comply with Chinese laws but the model is open source, you don't need to use their API when it can be self-hosted or hosted elsewhere without giving them any of your data or indeed being subjected to whatever additional censorship they may be obliged to add to their API. People absolutely can use it (and indeed are doing so).
It’s not a fact though - there’s no evidence to support the claim. Like what are the chances that some small company just revolutionised model training against the likes of OpenAI, Google, Meta, etc. It’s going to take a couple months to really distill what’s happening here.
If they did create the model on their own, at a fraction of the cost, on lesser hardware, then that’s great news. But if the just copied someone’s homework, refined it to target benchmarks, had extra government cash injection, etc. then it’s nothing new.
Soldato
- Joined
- 17 Oct 2002
- Posts
- 3,088
You can get output that rivals ChatGPT, and it can run locally on a 6GB laptop GPU - that isn't a big deal to you?
It depends on the model itself. If the model pre-trained, can generate the problems then it will still remain. For example, if the LLM can generate output that has vulnerability because the training data is biased or has references embedded (ie the weighting creates a text string that outputs bad code).
If the open source requires you to train your model then it's nothing more than an empty husk and I agree if the T&C support the use. However users are still assuming it's great and the real world results may be a different story. However we shall see.
Not many users can take an open source project, adopt it then maintain and improve it. So speed of advancements are something to consider the ROI.
The release was designed to offset the US and Trump however a large portion is blown up by the US media.
The model weights are also available. I expect the while model to be on AWS bedrock within a few months znd probably z guide to deploy it on Sagemaker within dzyd
I don't think the perceived threat for the market is this particular model it is the clear evidence that highly functional AI can be developed on much smaller budgets and trained using much cheaper hardware that has spooked the financial markets. People are betting huge sums of money on AI giving them huge profits down the line which just won't be there if companies are able to undercut development costs by multiple billions.
DeepSeek is trained using other LLMs, so without the billions spent by other companies, it wouldn't exist. It's an interesting solution for more targeted models, but it doesn't undermine what's already been built.
Nvidia stock price says otherwise? Nearly 600bn wiped
blind leading the blind.
did we see nuclear power investors disappear due to lack of needing more nuclear reactors? No..
Yea because the stock market is driven purely by fundamentals. People always panic, and it causes a chain reaction.
Anybody thinking we suddenly don't need these cards is mental. If anything, this makes building LLMs more accessible i.e. more GPUs.
Its also probably a releveling. ChatGPT burst onto the scene and everyone related to it got a massive share price boost. Competitors emerge and that is probably a realigning with reality. I expect Nvidia to drop further overtime, but still way above where they were before ChatGPT showed up.
I expect them to drop over time if other companies are able to produce comparable hardware and the necessary tooling to take advantage of it. There are some on the horizon but this feels like a speedbump that will correct itself.
24% drop .. in the end they will still get their trade because it won’t stop at one AI.
Aws and google are making their own hardware so it’s natural.
Associate
- Joined
- 21 Oct 2002
- Posts
- 1,028
I think this is all expected. Market settling.
Yeah I don't think people understand this - there's lots of handwaving about censorship but it's not really sunk in that this is an open model, it doesn't have to be hosted in China, it doesn't require a censorship API, other third parties will host it (it's not just China vs do it yourself). Lots of buzz surrounding AI but plenty of people misunderstand it.