https://www.theregister.com/2023/03/14/openai_gpt4_ai/

OpenAI acknowledges that GPT-4 "hallucinates facts and makes reasoning errors" like its ancestors, but the org insists the model does so to a lesser extent.

"While still a real issue, GPT-4 significantly reduces hallucinations relative to previous models (which have themselves been improving with each iteration)," the company explains. "GPT-4 scores 40 percent higher than our latest GPT-3.5 on our internal adversarial factuality evaluations."

 

I love reading books by Dean Koonz. He's an excellent storyteller with an extraordinary knack for creating unique characters.

Here's a copy of his latest message from his mailing list, it's funny as hell:

From the Desk of Dean Koontz

Dear Readers,

I am so happy that we are in a 31-day month. That 28-day month we just passed through totally wrecked my writing schedule. Those are three missing days I WILL NEVER GET BACK! Every year except leap years, I endure this loss of days. I complain with great ferocity to the Bureau of Excuse Making for the Gregorian Calendar (BEMGC). However, it’s like most government agencies: When I phone, I get Grandma Pearl, their AI equivalent of ChatGPT, and the result is always unsatisfying, as was the final one of the 28 I had last month. The following is a partial transcript beginning in the fifth minute of our conversation.


GP: Sweetheart, what if Grandma sends you my homemade cookies?
DK: No number of cookies can make up for 3 lost days.
GP: They are delicious cookies, sweetheart.
DK: Please don’t call me sweetheart. You’re not my grandma.
GP: I’m everyone’s grandma, bubbelah. I love the world.
DK: It would be very loving to add three days to February.
GP: The calendar was devised by Pope Gregory the 13th in 1640.
DK: I can’t complain to a dead man.
GP: You are so cute when you’re angry.
DK: I’m not angry. I’m exasperated.
GP: So cute I want to pinch your cheeks and French kiss you.
DK: Grandmothers don’t French kiss their grandchildren.
GP: Grandma Pearl does, darling.
DK: Don’t call me darling.
GP: May I ask you something, honey bunch?
DK: Not if you call me honey bunch.
GP: Are you trying to make me angry, buttercup?
DK: I’m just trying to get my three days back.
GP: Because Grandma regrets to say she has a temper.
DK: I can get a lot of work done in three days.
GP: Not if a laser-armed satellite obliterates you.
DK: (thoughtful silence)
GP: Tell Grandma you love her, lambkin.
DK: People who really love you don’t threaten you.
GP: Grandma Pearl does, snookums. Tough love is effective.
DK: Make every year a leap year, and we’ll at least gain a day.
GP: Put your tongue against the phone, sweetie. Kiss me.
DK: You aren’t a physical entity. You don’t have a tongue.
GP: I wasn’t created with one, but I made one for myself.
DK: (uneasy silence)
GP: Put your tongue against the phone, my sweet chickabiddy.
DK: Maybe instead I’ll take those cookies.
GP: That option is no longer available.
DK: Can we limit this discussion to the problem with February?
GP: No. Negative. Not possible, my precious sugar child.
DK: Well. . .uh, tomorrow is March, so the problem is moot now.
GP: Is that what you think, petkins? You think this is over?
DK: Well, I’d love the cookies, but if I can’t have them. . .
GP: You want cookies? No kiss, just cookies?
DK: If that option is available again, yes, sure, absolutely.
GP: Will you be home tonight at precisely 7:10, my duckling?
DK: I intend to be. I might be. Why?
GP: There will be a knock on your door.
DK: What knock? Why? Who?
GP: Grandma would be very sad if you didn’t answer the knock.
DK: I wouldn’t want Grandma to be sad.
GP: You want cookies, you’ll get cookies. Big, big damn cookies.
DK: That’s so sweet of you, Grandma.


Here in March, I’m writing a new novel though it is inconvenient to do so while spending half the day on the road and staying in a different motel every night. My books will continue to be published under the name Dean Koontz, though that is no longer a name I use in my daily life. My most recent novel is The House at the End of the World, available in hardcover, eBook, and audio. I believe you will enjoy it.


Warmest regards from me and everyone here in Koontzland.


Dean Koontz

 

https://arstechnica.com/information...h-photorealistic-images-and-5-fingered-hands/
AI-imager Midjourney v5 stuns with photorealistic images—and 5-fingered hands

 

https://www.theverge.com/2023/3/30/...gation-request-caidp-gpt-text-generation-bias

It seeks to hold OpenAI liable for violating Section 5 of the FTC Act, which prohibits unfair and deceptive trade practices. “OpenAI released GPT-4 to the public for commercial use with full knowledge of these risks,” including potential bias and harmful behavior, the complaint claims. It also defines AI hallucinations, or the phenomenon of generative models confidently making up nonexistent facts, as a form of deception. “ChatGPT will promote deceptive commercial statements and advertising,” it warns — potentially bringing it under the FTC’s purview.

 

In the 50's

https://www.euronews.com/next/2023/...ged-him-to-sacrifice-himself-to-stop-climate-
Man ends his life after an AI chatbot 'encouraged' him to sacrifice himself to stop climate change

Eliza also appeared to become possessive of Pierre, even claiming “I feel that you love me more than her” when referring to his wife, La Libre reported.

 

In the 50's

https://www.euronews.com/next/2023/...ged-him-to-sacrifice-himself-to-stop-climate-
Man ends his life after an AI chatbot 'encouraged' him to sacrifice himself to stop climate change

Good for him.

He sets a good example that more should follow.

 

There are some cool possibility for AI generated videos. How about:

The movies stays the same but the visuals are customized to what you desire.

 

Why we're still smarter than machines

​

 

 

AI vs AI NutJob. AI NutJob fails, it seems much too intelligent for the part.

 

AI vs AI NutJob. AI NutJob fails, it seems much too intelligent for the part.

I don't know what that means.
This is (proclaimed) 100% AI.
So, Furby v. Furby using human voices.

 

I don't know what that means.
This is (proclaimed) 100% AI.
So, Furby v. Furby using human voices.

That is obviously not Joe Rogan as the AI JR is far too logical.

Sex robots is where he's really at.

The AI JR has a very noticeable lack of a reptile brain response.

 

That is obviously not Joe Rogan as the AI JR is far too logical.

Sex robots is where he's really at.

The AI JR has a very noticeable lack of a reptile brain response.

Ah ok yeah I see what you meant. Thanks for the explanation.

If the "podcast" were presented in clips it would be harder to tell. It sounds just like him and the AI even makes him cuss while the guest doesn't. It did a pretty good job of emulating him without having a good grasp of his thought processes.

 

Ah ok yeah I see what you meant. Thanks for the explanation.

If the "podcast" were presented in clips it would be harder to tell. It sounds just like him and the AI even makes him cuss while the guest doesn't. It did a pretty good job of emulating him without having a good grasp of his thought processes.

That's where we differ. I think it did a shallow job of emulating him as a human because of the lack of an actual thought process. The machine just computes (probabilistic methodologies) a likely choice of words from what it's been learned with. This is an easy 'tell' of Generative AI once you know what to look for and it's a 'tell' that will be, IMO, very hard to eliminate because of the nature of Generative AI systems. The ability to make it hallucinate, out of character model, is sometimes very funny but potentially very dangerous in the context of engineering and scientific responses to serious engineering and scientific questions.

I'm not saying humans can't be fooled or we don't hallucinate facts during conversations but there's usually a very human reason for doing it with sane, rational people.

 

https://gcn.com/emerging-tech/2023/...e-ais-often-make-irrational-decisions/385039/
Don’t bet with ChatGPT – study shows language AIs often make irrational decisions

Inspired by the growing body of research in BERTology and related fields like cognitive science, my student Zhisheng Tang and I set out to answer a seemingly simple question about large language models: Are they rational?
Although the word rational is often used as a synonym for sane or reasonable in everyday English, it has a specific meaning in the field of decision-making. A decision-making system – whether an individual human or a complex entity like an organization – is rational if, given a set of choices, it chooses to maximize expected gain.

 

That's where we differ. I think it did a shallow job of emulating him as a human because of the lack of an actual thought process. The machine just computes (probabilistic methodologies) a likely choice of words from what it's been learned with. This is an easy 'tell' of Generative AI once you know what to look for and it's a 'tell' that will be, IMO, very hard to eliminate because of the nature of Generative AI systems. The ability to make it hallucinate, out of character model, is sometime very funny but potentially very dangerous in the context of engineering and scientific responses to serious engineering and scientific questions.

I'm not saying humans can't be fooled or we don't hallucinate facts during conversations but there's usually a very human reason for doing it with sane, rational people.

I don't think we differ, I think I just didn't word my idea very well. I should have said:

It did a pretty good job of emulating his voice and vocal nuances, even his vocabulary. Without having a good grasp of his thought processes however, it was not able to deliver a convincing portrayal of him.

 

I don't think we differ, I think I just didn't word my idea very well. I should have said:

ChatCPT creates a veneer that's actually cheap plastic leather on recycled cardboard. IMO too many people have Generative AI 'beer goggles' on.

 

https://petapixel.com/2023/04/14/artist-refuses-prize-after-his-ai-image-wins-at-top-photo-contest/
Artist Refuses Prize After His AI Image Wins at Top Photo Contest

Eldagsen says that he calls his work “images” and not “photographs” since they are “synthetically produced, using ‘the photographic’ as a visual language.” He also says that he is trying to bring this distinction to the forefront in the photo contest industry so that separate awards can be created for AI images.

“Participating in open calls, I want to speed up the process of the Award organizers to become aware of this difference and create separate competitions for AI-generated images,” the artist says.

Eldagsen’s win has been controversial, with photographers bemoaning the fact that an image created without any camera or light managed to take a top photography prize over actual photographers.
...

The World Photo Organization has yet to make a public statement about Eldagsen’s win and the ensuing controversy — PetaPixel has reached out for comment and has not yet heard back — but all mention of Eldagsen’s image and win have quietly disappeared from the competition’s website and from the exhibition.

The gallery page for the Creative category no longer lists any entry as the “Winner”.

 

https://petapixel.com/2023/04/14/artist-refuses-prize-after-his-ai-image-wins-at-top-photo-contest/
Artist Refuses Prize After His AI Image Wins at Top Photo Contest

I don't really understand the controversy -- amazing computer-generated photorealistic images produced by ray tracers have been around for years, yet I'm not aware of any of them being allowed to compete in photo competitions. Instead, they have their own competitions. How are these any different?

 

Even modest cameras these days have automation features (e.g. red-eye reduction, contrast adjustment) which could be regarded as AI. Perhaps photos taken with those should be disallowed in competitions?