Following on this recognition that the I in AI is Imagination, not Intelligence.
People are saying that text with typos is more likely to be human instead of AI.
First, specific shibboleths are doing more damage than good. The witch hunters are unfairly shaming or blocking real people. This gives the advantage to AI, as intended.
Second, every shibboleth is USELESS when dealing with a shape-shifting enemy like Altman. When OpenAI sees that people are judging by em-dashes, OpenAI will stop using em-dashes. That’s how it works, for fuck’s sake. Advantage to AI, as intended.
Third, if you take typographical error literally, ChatGPT’s pictures are NOTHING BUT typographical errors. Bad typography as well as bad spelling. The things that look like letters are not letters. The things that look like words are not words. The individual characters are weird dream-like mishmashes of several different letters, impossible to form with real lead type or digital fonts.
This must be deliberate. Computers have been doing excellent OCR for 40 years. OCR makes mistakes when the letters are distorted or the background is too complex, but it’s nearly perfect with clear text on a plain surface. The letters determined by OCR can be printed in an image easily. OCR never generates weird mishmashes; it sometimes mistakenly decides that a symbol is a Korean or Arabic or Chinese character, but it always prints a Unicode character.
These facilities are available to any ordinary computer. Altman’s machine COULD use the best OCR and the best graphics if he wanted to. He doesn’t want to do it.
I don’t see the advantage in these dream-like texts, but I’m sure Altman knows what it is.
Polistra's Mill 