Hacker News Reader

Well, thanks for not calling it open source! I did run it locally. It behaved as you would expect when asked about things the CCP cares about.

Very easy to circumvent if you run the raw model locally, no different from how western models are lobotomized for liberalist/nationalist reasons.

In fact these models are less lobotomized than western open source models.

Well, it is technically open source, open weights, and closed training set, right?

(My recollection is that the training code is MIT licensed.)

I don't see any training code in the GH repos. There is inference code for DeepSeek v3.

So "open weights" is accurate. We used to call this closed source...