Well, if we’re gonna be that pedantic about posts without sources to back it up, you’re not much better than the claim in question.
Your sidebar states
Share screenshots, analyze arguments, and discuss where they go wrong. Whether it’s science denial, historical revisionism, or bad math, let’s break it down!
The only thing happening here is a screenshot being shared.
The term ‘GPU’ appears 0 times in the linked document. Nor can I find any other sources of data that cite what % of GPUs are used to control racism. Keep in mind there are other techniques to control output, like careful curation of training data. Do you have any sources you are bringing? As far as I am concerned, the OP is yet another bit of imaginary anecdata, the typical fare of /pol/.
The impression OP provides is that the models will naturally emit racist material and it takes active effort to suppress it, which I don’t believe is the truth. As another tidbit, one is able to run local models and 100% of the effort goes into predicting the next token, not suppressing racist terms from being generated.
Your callout is fair but until you provide evidence to the contrary I do not take OP at face value.
Sure. Let’s take it at face value. OP said “a good portion”. What do you think a reasonable amount for good portion is?
Well, if we’re gonna be that pedantic about posts without sources to back it up, you’re not much better than the claim in question.
Your sidebar states
The only thing happening here is a screenshot being shared.
The term ‘GPU’ appears 0 times in the linked document. Nor can I find any other sources of data that cite what % of GPUs are used to control racism. Keep in mind there are other techniques to control output, like careful curation of training data. Do you have any sources you are bringing? As far as I am concerned, the OP is yet another bit of imaginary anecdata, the typical fare of /pol/.
The impression OP provides is that the models will naturally emit racist material and it takes active effort to suppress it, which I don’t believe is the truth. As another tidbit, one is able to run local models and 100% of the effort goes into predicting the next token, not suppressing racist terms from being generated.
Your callout is fair but until you provide evidence to the contrary I do not take OP at face value.