Suggestions on Detecting AI in Pseudo-News Quasi-Commentary
Dear All,
** Note for ease of consumption and simplicity I am calling todays GPTs "AI" even though they are more like predictive autocorrects.
As many might know I am a former NYC middle school teacher and current graduate student training undergrads. This has afforded me some experience in working with AI/generative slop, which has infiltrated this message board and continues to be difficult for some to recognize. I am here to share a few suggestions for AI detection.
1 -- The Cadance
While listening, if you hear an unusual prosody - a voice rising when it should normally stay steady, or a stacado when there should be a sing-song, you are likely listening to an AI. This is because it is not intelligence, it is simply producing content that it predicts is likely to follow given a prompt. It lacks the "right brain" capacity to make conversation have nuance and rhythm. To compensate, it randomly or semi-randomly incorporates human speech patterns, but these are rarely applied correctly.
2 -- The Gesture
If you are watching video, and you notice a gesture occuring often, it is likely AI. Sure, people gesture when they talk, but it is usually done subconsciously and if used rhetorically, becomes so associated with the speakers style that it stands out (consider Trump's hand accordion or Clinton's thumb/index finger emphasis). While watching AI, these gestures are performed at regular intervals - say, every 50 words. If you can see the gestures happening regularly, you likely are watching AI.
3 -- The Dichotomy
AI loves to simplify concepts into categories that obfuscate and oversimplify for ease of consumption. As such, when AI is used to make comparisons it does two significant things. One, it applies an equivalence (ChatGPT more than Grok) and then generalizes in an unuseful yet cognitively appealing way. "Who is a better superhero" you may ask, and the result you receive will be "There are many opinions on this so it is impossible to say. However, going by the numbers "Superman is an everyman with a better powerset, while Batman is the archetype of the Byronic dark hero who always wins the day". If you are forced to choose between a dichotomy you did not ask for, and the alignment seems arbitrary and oversimple, you likely have an AI.
4 -- The Hallucination
If you have provided a prompt seeking information, and you have clearly indicated a preferred result (i.e. "That storm was awful, were there any survivors?"
, you must remember the AI is not interested in providing factual information. It is interested in providing you with the most likely words that will follow that prompt, constrained by code that makes it tell you what you want to hear. In leading prompts, it will create information out of ether and unicorn farts, and the only defense is fact checking your sources. ChatGPT is better at this, because it at least shows you where it got its information.
5 -- The Glossy Aesthetic
AI constructed videos have an unusual glossy aesthetic that I can only describe as Pixar crossed with the old Rudolf stop action cartoons. If you see video that is glossy, shiny, and moves in an overly smooth manner given the budget of the production, suspect AI.
6 -- The Sigils
AI still has not mastered generating text that is secondary to its prompted mission. If AI is asked to write a story and include a picture, say, you may get a very good written story but take a look at the picture. Are there are words in the picture? If you prompted an AI to show you an image of library, read the titles of the books it generated (if possible). You will almost certainly note that there are alien symbols and weird glyphs in place of where actual words should be.
7 -- The Fingers and Teeth and Toes*
Still ain't perfect there. (*Source: Alice Kramden). Frequently there will be repeated or unclear fingers, toes, or teeth. AI is getting better at, but has not perfected, the fine details of image generation.
8 -- The Disclaimer
AI generated content is often flagged as "artificially constructed or altered content" in the information about the video. This is also often found on the thumbnail before you start watching this video. By design, this is usually harm to see but can often be found in a corner. If you see the phrase "content has been artificially altered" it is AI, no questions asked.*
*TexasTowelie for the source
What do you all think?
= new reply since forum marked as read
