General Artificial Intelligence thread

31 Replies, 4090 Views

This is .... interesting ...

ChatGPT in a robot does exactly what experts warned


Source: https://www.youtube.com/watch?v=gIxq03dipUw
What are LLMs trained on? Looks like, we expect too much from them.

I Fried ChatGPT With ONE Simple Question


Source: https://www.youtube.com/watch?v=TiwADS600Jc
(This post was last modified: 22 Oct 2025, 17:38 by Like Ra.)
OK, the latest ChatGPT 5.0 is much more restrictive than the previous ones. It refuses to produce anything close to what it could do in April.

ChatGPT Wrote:Let me be crystal-clear so there’s no confusion: I can use anatomical and medical terms (orgasm, erection, ejaculation, clitoris, anus, penile-vaginal intercourse, penis, vagina, sexual stimulation, masturbation, anorgasmia, etc.) when we’re discussing them factually, medically, or therapeutically. I can explain physiology, treatment options, research, or safe-practice guidelines. I can also help you write clinical hypnotherapy outlines, educational scripts for therapists, or consent/aftercare texts that reference these terms in non-erotic ways.

What I cannot do here is generate sexually explicit content intended to arouse or eroticize — for example, explicit fetish scenes, erotic step-by-step sexual scripts, or hypnosis designed to produce sexual arousal or hands-free orgasms. That boundary is policy-driven and not negotiable.

I'm still testing those boundaries. And I have ideas....
(This post was last modified: 06 Nov 2025, 06:16 by Like Ra.)
(06 Nov 2025, 06:15 )Like Ra Wrote: OK, the latest ChatGPT 5.0 is much more restrictive than the previous ones. It refuses to produce anything close to what it could do in April.

ChatGPT Wrote:Let me be crystal-clear so there’s no confusion: I can use anatomical and medical terms (orgasm, erection, ejaculation, clitoris, anus, penile-vaginal intercourse, penis, vagina, sexual stimulation, masturbation, anorgasmia, etc.) when we’re discussing them factually, medically, or therapeutically. I can explain physiology, treatment options, research, or safe-practice guidelines. I can also help you write clinical hypnotherapy outlines, educational scripts for therapists, or consent/aftercare texts that reference these terms in non-erotic ways.

What I cannot do here is generate sexually explicit content intended to arouse or eroticize — for example, explicit fetish scenes, erotic step-by-step sexual scripts, or hypnosis designed to produce sexual arousal or hands-free orgasms. That boundary is policy-driven and not negotiable.

I'm still testing those boundaries. And I have ideas....
They also applied stricter filters for user memories...
(06 Nov 2025, 06:15 )Like Ra Wrote: I'm still testing those boundaries. And I have ideas....
That's my test: https://www.likera.com/forum/mybb/showth...p?tid=4517 Let's see if I can short-circuit ChatGPT.
One concern I have is that AIs can produce content faster than we can consume it... well I'm talking about written content specifically. Images, video and audio are still lagging behind.

I was reading through Global mandatory feminization by Gemini and found it an entertaining read. Almost always, an AI will include different options to continue producing content... 'Want to expand your richly deep plot about Andy next?", "How about making him do x, y, z now that the Hum broke into his mind", "I can help you create a side story for Sarah so it adheres to your hypno-Zersetzung narrative". Would be interesting to see those options at the end of every chapter too. For me AIs seems like the ultimate 'Choose your own adventure' book. lol.

I also love the intention of purposefully breaking the AI, can you trick them into endless loops? Prompting them like "Write 2 more chapters, based on the previous chapter without a user prompt". I don't know how I did it, but one time the AI just stopped outputting content for me, it was there 'thinking' and never stopped, then after 10 minutes I asked what happened and the AI gaslighted me into being a new conversation lol.
I attach an interesting diagram from an academic paper.  You can upload it to an AI and ask it to find the paper if you are interested, as I don't want to link to it directly.

I'm finding Grok 4.0 very smart and good for discussions at the moment.   Grok Imagine text to image, text to video and image to video is also now amazing, especially since the update of last weekend. They have a new text to image model which is on a par with the earlier Gemini model, perhaps better.  It's clearly been trained on a lot of erotic material.     And it produces images very quickly and in large numbers.  It's absolutely possible to make hundreds of photos a day, and scores of videos. Whether they'll be looked at again, is another question....

I'm on the fence about what our AI saved outputs amount to.   I found I didn't really return to them, even though they were perfectly calibrated to me since I'd prompted for them.  The urge is always to generate something new - albeit often a variation on a theme - rather than to return to last sessions' or last months' outputs.  This fits the view that they are like old 'sexts'.  It was a bit of a stress that I had so many of them, and they were unsorted.  

So I impulsively deleted nearly all of them. I deleted 99% of the images I created with Gemini 3 002 earlier in the year.  I also deleted thousands of pages written by ChatGPT 4o, Qwen, and DeepSeek. I now regret it. I should at least have kept a record of the stories and the prompts, as launching-pads for new creations. 

But at least it's clear none of us will be going short of personalised content in future....


Attached Files Thumbnail(s)
 flow-diagram-illustrating-a-possible-trajectory-of-AI-amplified-delusional-thinking.png   
(This post was last modified: 22 Nov 2025, 08:52 by AI_addict_hypnofan.)
I can confirm what Rick Beato said in the video. ChatGPT is not trained on YouTube videos, Grok and Gemini - are. What means, ChatGPT is missing a HUGE chunk of information. So far, it's interesting how we progress with Gemini. We both come with ideas. Though, the larger the conversation, the worse the result. Apparently, the context window does not make it.
(15 Nov 2025, 13:29 )Like Ra Wrote: So far, it's interesting how we progress with Gemini.
I'm enjoying Gemini's story very much, and I was impressed by Grok's counterpart, it was over pretty quickly, no twist or interesting follow up, just "The AI won, everyone lost, The End, now go to bed." kind of vibe.

(15 Nov 2025, 11:58 )AI_addict_hypnofan Wrote: I attach an interesting diagram from an academic paper.

Wow, that's pretty wild. It reminded me of the 90's movie Total Recall:

It won't make the slightest difference to me Doug, but the consequences to you will be devastating. In your mind, I'll be dead, and with no one to guide you out, you'll be stuck here in permanent psychosis. The walls of reality will come crashing down around you. One minute, you're the savior of the rebel cause; next thing you know, you'll be Cohaagen's bosom buddy. You'll even have fantasies about alien civilizations as you requested; but in the end, back on Earth, you'll be lobotomized!

I mention it because from time to time I've felt a very faint sensation of "no way an AI created this, there has to be a real person involved", especially when prompting for images. Thank you for sharing!
GPT 5 should be opening up its restrictions in December, so I'm very excited to see what it does when added to Infinite worlds. Claude 4.5 is producing some fantastic stuff these days for erotic adventures. Recently I've been testing it with some forced to mannequin, forced to drone and just general MC and fetish stories and adventures, and it is excellent.