#my theory: ai detector is another genai but instead of generating text it generates probability matrix and compares it with the text sample | Explore Tumblr posts and blogs

superconductivebean · 1 month ago

Text

#1622

Quillbot introduced a new feature: level of confidence its AI-for-AI supposes something has been generated or touched by the AI.

This feature, as predicted, rolls like the utter shitewheel it certainly is, because of course the AI will mark M-dashes and parenthesis incised into the sentence as at least AI-refined. And no, there would not be a clear guideline for how to use this feature.

By that I mean a clear instruction for how to… *sigh* Text generation is math. A lot of teeth-cracking math, some rigid programming, and all of that is based in linguistics. Neither field is particularly easy to get into and have a solid grasp on, so it should be logical a tool presented as easy-to-use and entirely-comprehensible in regards of all this—so in the form of percentages and now Confidence—*must* include explanations for either category with which it will mark texts.

#днявочка #eng tag #and you see. ai detection software *does not want* anyone to know how it works. otherwise it might help refine genai #that and the desire to make money out of semi-false promises do not coexist #tldr; if we cant peek under the hood—we must reverse-engineer everything.#both ais are algorithms. it's not impossible to crack their logic #my theory: ai detector is another genai but instead of generating text it generates probability matrix and compares it with the text sample #you see where this is going. llms are plenty so if you do not know which of them generated the sample you can get false negatives #same with human writing. detector simply gets confused. ok listen it should be another post i'll type it up soon

3 notes · View notes