Yay another LLM! That’s definitely what the world needs and don’t let anyone make you think otherwise. This is so fun guys. Let’s fund the surveillance, stealing, misinformation, harmful biases, and destruction of the planet. I can’t believe some people think that humanity is more important than another “open source” crazy pro max ultra 8K AI 9999!
Well… 🤔
DeepSeek imposes similar restrictions, but only on their website. You can self-host and then enjoy relatively truthful (as truthful as a bullshit generator can be) answers about both Tianmen Square, Palestine, and South Africa (something American-made bullshit generators apparently like making up, to appease their corporate overlords or conspiracy theorists respectively).
Nope, Self hosted deepseek 8b thinking and distilled variants still clam up about Tianmen Square
If you’re talking about the distillations, AFAIK they take somebody else’s model and run it through their (actually open-source) distiller. I tried a couple of those models because I was curious. The distilled Qwen model is cagey about Tianmen Square, but Qwen was made by Alibaba. The distillation of a US-made model did not have this problem.
(Edit: we’re talking about these distillations, right? If somebody else ran a test and posted it online, I’m not privy to it.)
I don’t have enough RAM to run the full DeepSeek R1, but AFAIK it doesn’t have this problem. Maybe it does.
In case it isn’t clear, BTW, I do despise LLMs and AI in general. The biggest issue with their lies (leaving aside every other issue with them for a moment) isn’t the glaringly obvious stuff. Not Tianmen Square, and certainly not the “it’s woke!” complaints about generating images of black founding fathers. The worst lies are the subtle and insidious little details like agreeableness - trying to get people to spend a little more time with them, which apparently turns once-reasonable people into members of micro-cults. Like cults, perhaps, spme skeptics think they can join in and not fall for the BS… And then they do.
All four students had by now joined their chosen groups… Hugh had completely disappeared into a nine-week Arica training seminar; he was incommunicado and had mumbled something before he left about “how my energy has moved beyond academia.”
That’s not how distillation works if I understand what you’re trying to explain.
If you distill model A to a smaller model, you just get a smaller version of model A with the same approximate distribution curve of parameters, but fewer of them. You can’t distill Llama into Deepseek R1.
I’ve been able to run distillations of Deepseek R1 up to 70B, and they’re all censored still. There is a version of Deepseek R1 “patched” with western values called R1-1776 that will answer topics censored by the Chinese government, however.
I’ve been able to run distillations of Deepseek R1 up to 70B
Where do you find those?
There is a version of Deepseek R1 “patched” with western values called R1-1776 that will answer topics censored by the Chinese government, however.
Thank you for mentioning this, as I finally confronted my own preconceptions and actually found an article by Perplexity that demonstrated R1 itself has demonstrable pro-China bias.
Although Perplexity’s own description should cause anybody who understands the nature of LLMs to pause. They describe it in their header as a
version of the DeepSeek-R1 model that has been post-trained to provide unbiased, accurate, and factual information.
That’s a bold (read: bullshit) statement, considering the only altered its biases on China. I wouldn’t consider the original model to be unbiased either, but apparently perplexity is giving them a pass on everything else. I guess it’s part of the grand corporate lie that claims “AI is unbiased,” a delusion that perplexity needs to maintain.
What exactly makes this more “open source” than DeepSeek? The linked page doesn’t make that particularly clear.
DeepSeek doesn’t release their training data (but they release a hell of a lot of other stuff), and I think that’s about as “open” as these companies can get before they risk running afoul of copyright issues. Since you can’t compile the model from scratch, it’s not really open source. It’s just freeware. But that’s true for both models, as far as I can tell.
Yup, this is open weights just like DeepSeek. Open source should mean their source data is also openly available, but we all know companies won’t do that until they stop violating copyright to train these things.
I figured as much. Even this line…
M1’s capabilities are top-tier among open-source models
… is right above a chart that calls it “open-weight”.
I dislike the conflation of terms that the OSI has helped legitimize. Up until LLMs, nobody called binary blobs “open-source” just because they were compiled using open-source tooling. That would be ridiculous
Open weights + an OSI approved license is generally what is used to refer to models as open source. the with that said, Deepseek R1 is am MIT license, and this one is Apache 2. Technically that makes Deepseek less restrictive, but who knows.
