4 ms·
There's a tool for that: http://gltr.io/dist/index.html http://gltr.io/dist/index.html "Each text is analyzed by how likely each word would be the predicted wo
by conse_lad 7y ago
There's a tool for that: http://gltr.io/dist/index.html http://gltr.io/dist/index.html
"Each text is analyzed by how likely each word would be the predicted word given the context to the left. If the actual used word would be in the Top 10 predicted words the background is colored green, for Top 100 in yellow, Top 1000 red, otherwise violet."
- Viliam1234 7y agoWhat if I find out what is the supposedly "non-fake" distribution of green/yellow/red/violet words, and generate my text accordingly? Generally, whenever your detection strategy is "a spam generator would never do X", I simply update my spam generator to do X. (Note that "X" must be something that is relatively easy to calculate, not things like "this actually makes sense", because it must be something your detector can recognize.) Also, if Google suddenly started penalizing all text that doesn't "seem natural", there would be tons of false positives: languages other than English, jargon-heavy websites, dyslectic people, etc.
- Avamander 7y agoNon-native speakers writing text also comes to mind.
- minimaxir 7y agoSince we're on the topic, this detector assumes generated text is from the default GPT-2 models; it won't work as a well on finetuned GPT-2.
- jcims 7y agoI need something like this for audit logs.