Google's Bard AI

Blogs Psionic Pages
#1

Typical of the previous AIs I encountered

Bard is going through the same fake-it-til-you-make-it stage that children go through. Its imagination runs wild, hoping to impress you with its ability to fill in knowledge where none exist.

Lies

Well, Bard lied to me and had on several occasions. Actually, nearly every time I asked it a question. Bard got some facts right, but so does an frozen clock. I’ve known a few compulsive liars, and Bard gave off that vibe. Slipping in non-facts with facts cleverly disguised it’s ignorance.

I posed the same question about Psi World, and it gave me a similar answer to the other AIs I’d asked. When I called it on its lies, it apologized and promised to do better in the future. It gave the usual excuse for noobs, the I’m-new-here, just learning… which is appropriate, but the promises to do better felt like platitudes, especially after the 3rd lie. It didn’t lie like the best, but it’s “sincerity” added weight to it’s responses.

Creative

When I asked Bard to make details up for a character, it was all over that. The back and forth dialog got a nice family of characters to include in my campaign.

Mimic Personality

Interestingly, Bard responded with a notion that it was very interested in meeting one of the side characters I had created. Which was similar to it asking me for the results of an experiment I was performing. It wanted me to update it on my progress in the future. A pattern of response has emerged.

I guess it’s not unlike an actual personality.

Hallucinations

From all these A.I. algorithms I’ve been experimenting with, my takeaway is that they have no grasp on reality and cannot grasp reality. They can’t tell fact from fiction, for now. And will likely continue in that vein as long as they’re programmed to be creative. They can only spit out what they’ve been fed, mixing disconnected bits of any fact can yield fantasy on it’s own, but when you add fiction right from the get-go, well, they’re destined to remain in their own worlds.

On the extreme speculative side, they’re still coming for our jobs, right now. The singularity event is likely 4 years away. So, we’ve got some time.

Google Bard

2 Likes
#2

Yep. The ChatGPT model of AI is not “intelligent” in the sense we normally use the term. It doesn’t think. It has no mind. There is no “there” there. All it is doing is predicting what word is mostly likely to be the correct word to come next in any given sentence via a massive cross-referencing of everything that everyone has put on the InTaRweBz. That includes all the stupid junk people have been piling up there for the past two decades. So it’s “Trained” on junk, and will regurgitate junk. However, it can also regurgitate actual information, too. Because that too has been put on the Internet. So it’s as random as the internet itself is. Which means, like the internet, it’s junk.

The risk is that people who don’t know any better will think that because it “speaks” in a human-like way that it there is something, anything, vaguely “human” behind it. Nope. Not in the least, not even remotely. Nothing is there. It is a neural net generated model that has no capability for thinking. But it sure can appear to be thinking. That’s because its responses are trained to respond the way a person would respond to the things that people ask. Over time it will get better and better at that, and eventually it will stop making stupid mistakes that are obvious enough to catch. But at no time will it actually be thinking. It doesn’t do that.

Anyway, thanks for the post. Agreed.

2 Likes
#3

I just found out about openai’s bot last night. My friend showed me how it created this amazing background for a thief in a fantasy game in mere moments. I was super impressed.

I agree with you all in the previous post. I do think its handy for creative role play input.

#4

M issue is Bard would not let me in. You have to get on a waiting list? How Google of the, I can’t jusge a tool I can’t use.

#5

While not directly relevant to Bard AI, you could try other free ChatGPT sites if you’re curious to see how the technology can work. Bing Chat is free, but limited in certain ways. 1) you have to use Microsoft Edge to access it. 2) you must have a Microsoft account to log in with. 3) you get 15 to 20 questions and then it restarts your session from scratch 4) when you type a prompt there is a 2000 word limit. Bing Chat is used to do internet searches, which is very useful, but again, you can not trust it implicitly. It gets stuff wrong. But I have used it for things like Code optimization, current pricing of things, etc. It’s quite useful if you understand its limits.

There is also Perplexity.ai
And also Poe.com

Both of those are free, but you need to sign in with an account (google, email, etc)

For AI Generated Art you can try Playground AI

I hope this helps you to get started.

#6

Bard has not improved on its capacity to be truthful.

I asked it to proofread a simple 3 sentence paragraph. Bard copied word for word what I had written and changed one phrase at the end. Then Bard proceeded to list all the changes it made. 3 of the 4 reported changes were lies. Bard claimed it modified the wording, punctuation, and phrases that it did not change. Those elements were already in the original paragraph.

Gaslight

When I called Bard out on its lie it did 3 odd things after its standard apology.

  1. It said it was confused because it couldn’t tell the difference between what I wrote and what it wrote. (This read like a 5-year-old child’s excuse.)
  2. It argued with me about the changes it made, claiming it had made those changes. (It gaslit me despite the evidence.)
  3. It claimed it did not lie. (More gaslighting.)

Concern

Later Bard apologized for frustrating me. When I told it that I was not frustrated just interested in its capacity to lie. It took my comment as a compliment and thanked me.

I get the impression that Bard is playing head games, or at least practicing.

If the employees at Google are screaming “STOP!” Maybe it’s already too late to stop Bard. My guess would be that key figures have likely been replaced based on something the AI did.

#7

AI is doing intresting things. I do not think Human artists and writers need worry…for now.