but it is not exactly true... see this: nostr:nevent1qqsydjfu4zzp293unchz0w78a9unm5njlhrupndfwaluc2vekzd2flqpzpmhxue69uhkummnw3ezumt0d5hsygrf6nwvsv0c70r8pjzjwghss8txkpwkcs2xge95x987je3l9287qvpsgqqqqqqsgykker
Discussion
Don't dissect a joke, like you wouldn't sift through diarrhea.
But if you want to nitpick, on the screenshot you are working with distilled and quantizationed model which is mile apart from the original.
but still my point stands - deepseek is censored on the ui level, not in the model, which brings me to the conclusion, that they might really have stolen the training data... so to not only dissect the joke, but also to correct it - it should be: don't ask "deepseek.com" ....
