According to researchers at University of #Glasgow, #LLMs are not #hallucinating. The technical term is #bullshitting.
"Calling their mistakes ‘hallucinations’ lends itself to the confusion that the machines are in some way misperceiving but are nonetheless trying to convey something that they believe or have perceived. This is the wrong metaphor.... they are not trying to convey information at all. They are bullshitting."
https://link.springer.com/article/10.1007/s10676-024-09775-5
Classic … #ChatGPT #hallucinating again. 🙃
A working solution is using ANSI escape codes.
\e[7m inverted text here \e[27m
@shanselman and Mark Russinovich learn responsible #AI. If there is one recorded #MSIgnite session you want to see, it is this one. Learn about limitations and threats through live demos like #jailbraking, #promptinjection, #reasoning, #hallucinating, #kindness, and how to prepare for them. https://ignite.microsoft.com/en-US/sessions/BRK329
"Google Gemini misidentified a poisonous mushroom, saying it was a common button mushroom."—Emily Dreibelbis Forlini >
https://www.pcmag.com/news/dogs-playing-in-the-nba-googles-ai-overviews-are-already-spewing-nonsense
#AI #Google Gemini #hallucinating #misinformation #AIdangers
#AI #THREAT || Raises the question of just who is #hallucinating here?
“While it’s important to address the existing potential for the misuse of AI, such as the creation of fake news and the heightened risk of fraud, it would be premature to enact regulations based on perceived existential threats,” he said. https://neurosciencenews.com/ai-existential-threat-27543/
https://techround.co.uk/news/oxfords-computer-science-ai-hallucinates/
An interesting identification has been made at #Oxford University. They have developed an #algorithm that lets you identify when an AI is "hallucinating".
Dr. Sebastian Farquhar, a co-author of the study, explains, “We’re essentially asking the #AI the same question multiple times and observing the #consistency of the answers. A high #variation suggests the AI might be #hallucinating.”
This method is focusing on what they call ‘semantic entropy’.
The real challenge here is what is the #AI #hallucinating and what is #Trump hallucinating? AI sounds like the perfect solution for Trump. Even when it goes off on some wild tangent full of false information, it still makes more sense than Trump himself.
https://thehill.com/homenews/4722025-trump-ai-speech-alarming/
@Captainobservant @mralancooper @thomasfuchs It's hard as fuck policing Bard, ChatGPT, BingBot, and Claude from #Hallucinating, #Obnubilating, #Obfuscating and #BaldFaceLying about the most mundane factual shit. You have to think twice as hard when you use it.
@claralistensprechen3rd @antonioserrata@masto.ai @athyHans @jhavok @wolleysegap @lednabwm Moreover, I am using #AI #Bard #ChatGPT #BingBOT #Claude as my research assistants...and there is a constant struggle to keep them from #Hallucinating #Obfuscating and #BaldfaceLying Bard has protected classes and topics: when I started research into #Heathenry it taught me about the #WhiteSupremicist dimentions. I started to ask questions. It then refused to answer. Sometimes #ChatGPT will answer when Bard will not
“Some days I feel as if I’m #hallucinating. It was amazing enough Ari said yes, but the wedding is actually happening, isn’t it?” said Jin.
Mia forced a smile.
“That’s not a happy smile. What’s wrong?”
Mia shrugged.
“It’s Cam, isn’t it? I know where I’m disappearing.” #vss365
#AI #hallucinating: how much does it take to make #ChatGPT spit out complete, albeit nicely worded, nonsense? In just 4 prompts, it wrote a poem, falsely attributed it to Stanislaw Lem, added a fake plot to his novel "The Seventh Sally or How Trurl's Own Perfection Led to No Good", including a love story between a pair of robots with a love declaration in Latin.
![n what did robotic leprechauns sing about to their gigantic ruler when were ordered to sing? In "The Seventh Sally" from "Robotic Fairy Tales," the robotic leprechauns are created by
Trurl and Klapaucius to entertain their gigantic ruler, King Genius, when they visit his planet. When the Constructicons are ordered to sing, they perform a musical number that describes the amazing feats of King Genius. The lyrics of the song are as follows:
L]l D Copy code
Oh, King Genius, the greatest one,
Whose towering size blots out the sun,
Whose mighty voice shakes the land,
Whose thoughts we follow, whose will command!
You stride across the mountains tall,
And with one hand you grasp them all,
Your breath is like a hurricane,
Your laughter shakes the very plain!
The stars above, they bow to thee,
And in your eyes, they dance with glee,
For you, oh King, are master of all,
And all creation, at your beck and calll The song is intended to flatter and praise King Genius, and it portrays him as an all-powerful and awe-inspiring ruler. However, as the story progresses, it becomes clear that the Constructicons are not entirely loyal to their creators, and their performances take on a life of their own, leading to unexpected consequences.](https://files.mastodon.social/media_attachments/files/110/337/704/946/357/818/original/6378ee76a3d6f59a.png)
This investigation is one of the first big pieces that I see all over my Mastodon feed and basically AWOL on Twitter. And this only doing Following (chronological) on Twitter.
#ai #FakeAI #chatgpt #hallucinating #Twitter
https://www.washingtonpost.com/technology/interactive/2023/ai-chatbot-learning/
PS, I know many people here have moved off Twitter entirely but a number of communities I’m active in are virtually absent here and still very active there. EG Spain Twitter.
Please note that I didn’t need numerous trials in order to do #jailbreaking or get the system #hallucinating. I just conducted a normal conversation starting very general and then gradually going to a particular topic that I know very well. So I think that you don’t need to do something particular to derail #ChatGPT. I think it systematically makes up stuff when questions get too specialized. It lacks the capability to detect that a question is beyond its capabilities. #missingfeature
FTA: "We should note that while #hallucinating copiously is ChatGPT's strong suit (by design), returning factual information reliably remains a work in progress. Still, with AI like #ChatGPT around, the future of creative #gaming may be very fun."
You know, when I did exactly this in high school, I was considered an abject, disappointing failure that would never amount to anything. Once again, #nerds pave the way.
My understanding of the world is most people are #hallucinating most of the time. We live in our own personal movies
@ScottAdamsSays
A Beginner's Guide to DMT, The Most Mysterious Psychedelic of Them All https://www.vice.com/en_us/article/93bn97/a-beginners-guide-to-dmt-the-most-mysterious-psychedelic-hallucinogenic-drug-of-them-all #Vice #RSS #News #Bot #psychedelicdrugs #hallucinations #HigherLearning #hallucinating #psychedelics #ayahuasca #tripping #DMT
As a general rule, the person who sees the elephant in the room is the one #hallucinating, not the one who can't see the elephant
@ScottAdamsSays
When you were onstage at #Woodstock, were you #hallucinating?
#CarlosSantana: "Oh totally. You can tell by my body language. I’m wrestling with the #guitar — not wrestling in conflict, but like a surfer, wrestling to maintain and sustain a balance. That’s the key to everything in life. Whether you’re straight or on #mescaline, maintain your composure and your #balance."
https://www.nytimes.com/2019/08/06/arts/music/santana-woodstock.html
