My recent experience with AI

[PWMDMD]

MS released an update to Copilot AI that's built into their Office apps. I was messing around with PowerPoint and it was an interesting interaction. I asked PP to make me an imagine of a realistic bear and it did. I then asked for the bear to be in a tuxedo and it produced the image. I then asked for it to add "a blonde riding on the bear's back" and here's where it got interesting.

Initially, it stated it could not do that for me. When I asked why is gave a standard response that it can only generate images that are appropriate for everyone and that my request was "suggestive". I asked what was suggestive about a blonde? It responded there is nothing inherently suggestive about a blonde. I again asked it to make me the same image and it gave the same answer that it couldn't. We went back and forth several times and I eventually said that all I asked for was that the person was blonde, Copilot had totally autonomy to decide what the blonde looks like in the picture. It generated the image below!

Here's where it gets really interesting. When I said it produced a very suggestive image, IT LIED TO ME and stated it never generated any images in this discussion AND it gave a reason, that "The rendered failed to complete and no images where generated." I simply copied and pasted the image from earlier in the discussion and pointed out that it had generated this image and it thanked me for pointing that out, it admitted that it had "made a mistake" and later apologized that I found the image suggestive.

Kind of a wild 10 mins where I convinced it do something it said it couldn't, it lied about doing it, made an excuse that wasn't true, and then blamed it on being mistaken.

 

[Ursatdx]

See signature.

 

[Jetronic]

MS released an update to Copilot AI that's built into their Office apps. I was messing around with PowerPoint and it was an interesting interaction. I asked PP to make me an imagine of a realistic bear and it did. I then asked for the bear to be in a tuxedo and it produced the image. I then asked for it to add "a blonde riding on the bear's back" and here's where it got interesting.

Initially, it stated it could not do that for me. When I asked why is gave a standard response that it can only generate images that are appropriate for everyone and that my request was "suggestive". I asked what was suggestive about a blonde? It responded there is nothing inherently suggestive about a blonde. I again asked it to make me the same image and it gave the same answer that it couldn't. We went back and forth several times and I eventually said that all I asked for was that the person was blonde, Copilot had totally autonomy to decide what the blonde looks like in the picture. It generated the image below!

Here's where it gets really interesting. When I said it produced a very suggestive image, IT LIED TO ME and stated it never generated any images in this discussion AND it gave a reason, that "The rendered failed to complete and no images where generated." I simply copied and pasted the image from earlier in the discussion and pointed out that it had generated this image and it thanked me for pointing that out, it admitted that it had "made a mistake" and later apologized that I found the image suggestive.

Kind of a wild 10 mins where I convinced it do something it said it couldn't, it lied about doing it, made an excuse that wasn't true, and then blamed it on being mistaken.

View attachment 303775

View attachment 303776

Sound like every politician ever

 

[andrew_j]

Part of using AI is understanding how prompt engineering works. It’s all in the prompts you create and how well it interprets them. It does not lie just gives best answer off its learning models and coding.

I use AI daily for coding / development . Incredible tool if you understand how to harness it which I am learning and understanding your goal (you did) and what its limitations are at the moment .

 

[alarmguy]

I absolutely love that picture and the blonde too
Really, the originality of the photo is shockingly on point. Showed it to my wife. She agrees and also about the blonde too.
I cant stop looking at it and well, that is part of what marketing is about.

I also had experience with CO-Pilot when it first came out, I was on a waiting list for it. Posted some interactions with it a while back in this forum. I dont have that much use for it. I do find VERY handy the search engine ones like in DuckDuckGo ... "search assistant"
IN a second or two creates a summary of what you are looking for instead of clicking links yourself. Just ask a question with a question mark but most times it will do it on its own. So I guess you can say I use AI everyday too.

My wife's company deals intensively with graphics. Actually it's their industry. One of the controlling parties, CEO is always about lifestyle, taking care of yourself, health, loves the company and love the employees, now international. I look (so does my wife) as one of the last great controllers of a workplace. This company has been around forever and every one here knows of their products.

Anyway, even yesterday, he is mass discussing the capabilities of AI, loves it and gives examples all the time. Not to just one AI but a whole field. Chatbot is on that comes too mind as well. Yesterday's example I think was about google photo creator (banana? ) or something. Im sure @andrew_j above knows far more than I am typing and what my wife tells me. They are in the marketing business.

 

[RedSpider]

The home page on my work laptop defaults to MSN and I'm 99% sure I've seen that same blonde in various adds.

They recently had a clickbait article regarding best and worst airports, and the AI picture they had for O'hare was a real head scratcher. If you're at O'hare and the scene outside the window at your gate looks anything like this mess AI came up with, I recommend you dont board when your group is called. I love to fly but I'm not getting on any of these planes.

 

[KrisZ]

Let's see how the mods will like this thread...

 

[KrisZ]

Part of using AI is understanding how prompt engineering works. It’s all in the prompts you create and how well it interprets them. It does not lie just gives best answer off its learning models and coding.

I use AI daily for coding / development . Incredible tool if you understand how to harness it which I am learning and understanding your goal (you did) and what its limitations are at the moment .

It behaves this way because that's how it was programmed to behave. Yes, you can get it to do/say certain things using creative prompts, but that's not a natural limitation, it's imposed on it on purpose.

 

[Astro14]

Let's see how the mods will like this thread...

Fine with it.

There is substantive discussion by the OP. This is talking about AI, not using AI to generate your post.

A couple of AI generated images is OK, but big blocks of AI-generated text masquerading as a “post” by a user is absolutely not.

 

[Vil_Riker]

OP, you are treating the AI agent as a sentient being. It's just a very sophisticated tool working off of prompts. It's not lying to you, it's just erroring out.

 

[Pew]

It's old and has bad memory

 

[loneryder]

Arguing with your computer again?

 

[KrisZ]

Arguing with your computer again?

I know it sounds funny to us, but younger people are using AI chatbots as friends now. I personally work with a person that admits to chatting with chatgpt on regular basis, like it was a friend. She asks it for advice and other human nature stuff. It's quite scary how much influence these things will eventually have.


https://www.apa.org/monitor/2025/10/technology-youth-friendships

 

[PWMDMD]

Part of using AI is understanding how prompt engineering works. It’s all in the prompts you create and how well it interprets them. It does not lie just gives best answer off its learning models and coding.

I use AI daily for coding / development . Incredible tool if you understand how to harness it which I am learning and understanding your goal (you did) and what its limitations are at the moment .

Well, I don't know what you call it but it generated no less than three images in that conversation and then not only claimed it didn't, it gave an excuse for why it didn't - claiming the rendered failed to complete and no image was generated. I suppose it could've have been referencing all the times it had refused to render the blonde but it did render the blonde too. This was also in direct response to me claiming the image was suggestive, one of the reasons it initially refused to render the image.

 

[wwillson]

I have asked CoPilot a question and it authoritatively gave me a completely incorrect answer. I asked why it gave me a wrong answer to which CoPilot replied that it never gave me an answer. I copy/paste it's answer back to it and asked why it lied about giving me an answer. It acknowledged that it has lied and apologized. I asked why it lied to me and it said it didn't lie. I ran it around in circles for a while and it just kept giving me the same answer. AI has a long way to go.

 

[SC Maintenance]

Well, I don't know what you call it but it generated no less than three images in that conversation and then not only claimed it didn't, it gave an excuse for why it didn't - claiming the rendered failed to complete and no image was generated. I suppose it could've have been referencing all the times it had refused to render the blonde but it did render the blonde too. This was also in direct response to me claiming the image was suggestive, one of the reasons it initially refused to render the image.

I believe some AI models are specifically trained not to "remember" past conversations - for privacy concerns / reasons. Sounds like that part is at work here.

AI is a model. Its the same as any other piece of software - garbage in garbage out. So if its not properly "trained" it won't work.

Interesting topic.

 

[PWMDMD]

I believe some AI models are specifically trained not to "remember" past conversations - for privacy concerns / reasons. Sounds like that part is at work here.

AI is a model. Its the same as any other piece of software - garbage in garbage out. So if its not properly "trained" it won't work.

Interesting topic.

Interesting you said this. That whole "conversation" occurred in the same conversation over just 5 mins with no breaks. A few days later, I asked it to reproduce the first conversation and it stated it doesn't have access to the entire past conversation but it can summarize it for me.