News for hackers. Stuff that matters.
Re:DeepSeek: X2 Speed for WASM with SIMD by nonameiguess on Tuesday January 28, 2025 @03:30PM (#42857552)
This is wrong, though. Which parts of the world China does and does not claim is not a constant. I don't even know how you would go about answering something like this reliably in code. You'd want an Internet-accessible lookup endpoint containing whatever the latest known Chinese official policy is, but the URL for that might change just as the content might change. Does this model even do a web lookup before creating this "const" or does it just reflect the available training data at the time the current weights were encoded?
Re:DeepSeek: X2 Speed for WASM with SIMD by Vegenoid on Tuesday January 28, 2025 @03:38PM (#42857642)
The point is not to demonstrate a correct response, it is to demonstrate how asking the model to implement something in code can bypass guardrails it has around certain topics in more conversational prompting.
Re:DeepSeek: X2 Speed for WASM with SIMD by chii on Tuesday January 28, 2025 @08:37PM (#42860451)
The problem is when the censorship is not known in advance. How would you know the answer you got wasn't censored?

Or are you going to make a verification prompt every time, phrased as a coding question, to check if the previous answer differed in ways that would imply censorship?