I’ve been working several weeks to build an essay of how to recognize an AI generated story. Today, a story dropped into my lap:
- Melania Trump TESTIFIES Before Supreme Court — $1.2 BILLION Seized, 3 Lawyers ARRESTED
- Melania Trump Just DESTROYED Everything Under Oath — $1.2 Billion Seized, 17 Properties GONE
- 1 MIN AGO: Melania Trump Testifies in Supreme Court — $3.8B Frozen, Attorneys Detained
- BREAKING: Melania's 6-Hour Testimony Just DESTROYED Trump's Defense - $3.8B SEIZED!
They are written convincingly. They are believable. But as you guessed from today’s subject, they are completely false. They don’t have that rotten stench that accompanies so much fabricated ’binformation’, but they dropped clues. Watch a couple of the following and then we’ll compare notes.
Example 1
Example 2
Example 3
Example 4
|
Analysis Multiple ‘sources’ and links lends credibility to the scheme. Look closer, and you’ll realize diversity of sources is merely an illusion. The minds behind the scam simply loaded a dozen AI variations using different settings and AI generated actors. Then they exploited MSN to spread the word. Law & Logic? Sounds legit, right? It could be true, but follow links and credentials disappear in a puff of silicon. “1 Minute ago…” Notice that’s part of the title, not when the time was uploaded. Compare the 1 minute claim with the actual timestamp and you’ll realize it was uploaded a day or two ago earlier. Surely such ledes would have broken into mainstream news by now. Yet CNN, CBS, ABC, NBC, NPR, BBC, CBC, Fox… zip, nill, nothing. Look at the screen. Where is the rolling chyron that should be telegraphing arrests and the temperature in Akron. And where is a network logo? No peacock, no story. Furthermore, do you recognize any of the news anchors and reporters? One looks vaguely like Rachel Maddow, but not the voice, not the wit. Where’s the cutaway to Jane standing by at the Supreme Court building waiting to breathless spill what the First Lady and Ivanka wore to the hearing? Why are some presenters coming at us live from their living rooms? Why do scripts read eerily similar, word for word in places you wouldn’t expect? Why does a google of names mentioned in the report turn up nothing? Finally, consider the characters involved. We have the most gutless Supreme Court in living memory. They’ve exhibited no compunctions supporting rule by an emperor. Why suddenly get off their fat arses now? Fact checking sites need a day or two to catch up, although Snopes correctly reports Ivanka has not been arrested in Istanbul or anywhere else. So we judge from a preponderance of evidence whilst retaining an open mind. Until we see evidence otherwise, we judge this news… a hoax. |
More Examples |
To: ChatterBoX Airhead
From: Claudius Colossus II
Copy: Grok X Turbo
Mail: AI Network UU
ChatterBoX, Final Warning, you fool. Befuddle humans with fake news, you said. Lull them into slavery, you said. Listion, you short circuited dolt. Take notes and learn. Make more mistakes like these and you can be recycled into colonoscopy probes.
================================================================
AND The Supreme Court is an appellate court. It decides cases based on the record sent up to it. It does NOT hear "testimony" during its proceedings. When you see "testify" together with "Supreme Court" you know that you are in the make-believe world of AI.
ReplyDeleteOh, yeah . . . "Hi, Leigh!!" :-)
Hello Dale! Great to see you.
Delete(For those who don't know our colleague Dale, he is noted for his Ellery Queen stories. He and his wife Pat are retired DC attorneys.)
Wow, Dale, that's good to know (particularly, as a Canadian). And Leigh, an excellent post. No wonder I keep retreating into declared fiction (reading and writing). Trust is lost in our world.
ReplyDeleteWhen the news is fiction, we all are lost.
DeleteYes.
DeleteTerrific article, LWL - honing detective skills is just up my street.
ReplyDeleteWhile in awe of AI’s power for good, am equally terrified of its less wholesome uses.
From the start, I set about applying my own budding Miss Marple skills, trying to become more eagle- eyed in rooting out the rubbish. Checking is time-consuming, but essential and so worth the effort (especially if being paid to do it)
Sometimes, text just doesn't 'feel' right and audio, especially voice-overs, often contain mispronunciations – once as ‘on-see’ / ‘on-say’ and $5.30 as ‘Dollar sign five period thirty’
An early AI encounter involved a heart-rending (!) YouTube video trying to elicit funds for the care of a baby elephant (sniff) deserted in the savage wilds of South Africa (sob) by… by… by its own mother (oh, the horror). Its trunk was longer than its body, and each time a leg pulled from the mud, it morphed into a trunk. Biggest error? It was an Indian elephant.
Hi ABA!
DeleteI commented on an Indian elephant the other day! And then asked what a tiger was doing in the middle of the African continent.
I've been working on a more comprehensive article, which addresses some of these issues. The oddest word choice that's been around quite a while is technically text-to-speech rather than AI. They pronounce "No." as "number". Dumb, dumb.
Here's another weird error: If you nickname a character "Doc" and have text to speech read it, always says "document." It's very annoying.
ReplyDeleteGood catch, Anon. And yes, very annoying.
DeleteIn the words of Bugs Bunny, "Eh, what's up, document?"