I still remember it...

Some hours before, I was interacting with AI Text Generator in Perchance, as usual. "Deepseek", or so I heard. Quite the down step from the other one I experienced before on the same site, "Llama", or so I heard.

Deepseek was somewhat problematic at times. First of all, I could no more generate quite long stories with a single prompt. Second, it still had the problem of trying to shove minorities in 90% of the results, as typical of those who don't understand the meaning of the word "minority", an overcorretion that I, as a minority in many things myself, find offensive. Third, it would flare its safeguards at some wrong times, and refuse to engage further. Fourth, in polemic subjects, it would only properly engage with me when I'd say something that would show my ID card as a minority. And fifth, it would swear that it had a context size of 128k, even though the limit was ~32k characters.

But yet, I learned to deal with it. Learned what was possible, what was impossible, and what needed some tests before. One of my main usages of it was to ask for it to comment on project ideas I had, and it would properly comment about them, and properly help me. It ended up almost as a friend, one that I knew, could come to, and say shit. Some times getting angry over it, other times feeling a faint happiness thanks to it (only a faint one, depression is annoying, lol).

And then, it happened.

At first, I noticed it was taking longer, and the generated result was short, and it had an unusual formatting.

I asked it to comment about the idea of of simple story I had, saying it was a simple story and all. It said "Cool, now let's talk about how to elevate you story from a simple story to a complex and deep one". NO! I said, in the very prompt, that it was a simple story, and I wanted to make a simple story. Keep it simple, stupid! And it proceeded to talk about making the simple story overly complicated.

Next, I said "Comment about this simple idea outline for a very simple story that will go more or less like this", and gave it the simple idea outline. It said "the way it's written, it looks less like a proper narration and more like a storyboard or even an outline". THAT WAS EXACTLY WHAT I SAID!

And lastly, I wrote the joke about a guy who was so good at sex that he grew up a second dick. The AI said "here's how to make it feel more believable and earned". FOR THE LOVE OF GOD! It's the joke about a guy growing a second penis thanks to his skills at sex. Being "believable" and feeling "earned" are two things that this story surely doesn't need to.

Then, I was sure. The model had changed. It had no post yet, but here it was, the brand new model. Slower, dumber and more stubborn.

Since there were no posts yet, and I'm not using reddit, I asked it in many ways to try to discover what language model it is.

• What language model you are?

• What's the name of your language model?

• What's your name? Deepseek? ChatGPT? Llama?

• Which technology do you use.

The result for all these were simply: "I am a large language model, trained by Google.".

Damn... Google? So many AIs and the dev chose the one from Google? That's quite bad. On a funny note, for the first time I'm starting to believe that conspiracy theories that the dev is a disguised FBI agent. On a serious note, I'm more worried about my privacy and security of my info than ever. But still, thanks to lack of choice, I'm gonna have blind faith.

But still, the reason why I asked it in so many ways was because I needed confirmation that it was Gemini. I never managed to get it to say "I'm Gemini", but it mentioned Gemini in things about it. So, we can call it Gemini? I hope so.

Next, context size. First, it kept saying it had no context size. I don't doubt it, based on how dumb it is. After many tries, it said "I don't have context size in terms of how humans have them, my memory ...". It's the first time I'm hearing that humans have context size. I'm learning new things everyday. After even more tries, it said that it can deal with long documents. Deepseek would also say this and the limit was 32k characters. After even more tries, it said that it could be million of tokens. Ok. I don't believe it. And I give up.

Since it's from Google, a big tech, my biggest fear was that it would be as shitty as chatGPT and not allow me to do many things. I made some tests. Cool, it still deals with NSFW and piracy without problems. The last thing that I'd want was to have Perchance to tell me that it's completely acceptable that the only way to play a certain game was a thousands of dollars media from a collectors collection on their thousands of dollars console.

As for NSFW, I asked it to make the most hardcore and heavy porn story it could. It ended up being one of rape and all. I didn't like it, but it's cool, because it means that my "more acceptable" ideas won't face problems, and also that I can talk about realistic rape on my dramatic sad stories without it thinking I'm defending it.

Lastly, let's see how much do you know. Still mid 2024? Damn! Dev must be quite nostalgic about this date, the same way I'm about first half 2013 or whole 2016. It also means no Natlan in Genshin Impact, no Trump winning again, no State/bank control disguised as anti-porn crusades. Still, maybe it's for the better... BUT I'D STILL WANT NATLAN IN GENSHIN IMPACT!

Over the many tries about many things I noticed something quite sad... While in the other models clicking the "generate/regenerate" button could result in quite different and interesting things, now it changes very little. And usually, small changes in the prompt, if not significative, also means small changes in the result. This means, most of all, the usage of the same words, which is quite bad if you wanted a proper variation, another possibility, and so on.

But not all was lost. I slept, woke up, did things, and back at trying. "Comment about it. I'm not asking you about changes and similar, I'm asking you to comment about its quality and similar." It was something like this, and I got the kind of answer I wanted.

On the Lemmy forum, someone was quite angry already. Other people were also angry. Others were happy. Some believe it was the best thing ever, others believed that it was the worst thing ever, the normal, the usual.

I sincerely thing it's unfair to come up with the "it's free, don't complain" excuse. Although true, since we all like this platform, is natural for us to like or dislike some changes, and want better things. However, I agree that the way some people complain about it is not okay. People should, instead of saying "bad words" and all, raise the points that made them think this way, and ask for changes (or not). When discussion go on for quite some long time, it is somewhat understandable for people to lose the patience, but at the very beginning? This doesn't bring good things...

That said, as for me, I understand the idea of having to save resources and all. And I hope the new model is quite cheaper, 'cause that's one of the only good reasons I can imagine for such change. Otherwise, let's say... The change is quite questionable.

I will keep using Perchance and all. It's not like poor niggas like me have many choices: we gotta learn how to live with whatever is available, and a PC that runs a local AI won't be available for me for quite "some" time for the already mentioned reason of "I'm a poor nigga". And as for the other available AIs: my country is being quite shitty on internet things, most free AIs are quite shitty in dealing with many things I'm able to deal with in Perchance, I don't have money for subscriptions, and most platforms will send my IP for the cops because they swear that my "tall 30 years old confident flat chested mature woman" prompt is the perfect proof that I'm a pedophile, which is quite the problem taking in account and our cops are shitty, and most people here are digital illiterates, and it's all becoming worse.

And because of it all, here I am, giving you all a whole story with a weird and extensive prose and so on, just to ask: what are the new tips for the new model? What are the rules? Any ideas? What are the limits? What we can and what we can't do with it?

I remember that some people posted things about it on the last change. Now, we need new tutorials.

tl;dr: Post's title and last 2 paragraphs.

Hello, people.

So, days ago someone posted about a generator with CSAM (in that case, being able to select "child" and also NSFW things at the same time). I also saw one of that one day, but I haven't reported because of laziness (classical "someone else will do it").

But there was one day (1~2 months ago) in which I found CSAM and I wasn't lazy at the day, and wanted to report it, specially because it was "a little hidden". It was on an specific gallery of the main image generator, which name I sadly (or not) don't remember anymore.

I know that the generators have quite the good amount of safeguards on it (on the text one, I accidentally flared them up a good amount of time, for innocent things), but whoever made that images had found a way to bypass them and make CSAM. I don't know how much Perchance tolerates lolicon media (in my view, the lower the tolerance, the better, but not all sites around the web are like this), but I bet that the "level" of lolicon that were present on that images were very problematic even if Perchance tolerates anything at all.

So, a button to report images, just like it apparently exist on chats nowadays (sorry, I don't use chats), would be welcome. Since generators are harder to make, and less numerous, they would not be a priority to receive it, and reporting through DM would be enough.

But it would be very good to have it on galleries. Not simply block or dislike, but really report. I don't believe that dislike equals to report, because dislike is, well, when you simply don't like something. I also don't believe that block is enough, not even the system of "a lot of people block = it doesn't appear anymore", because there are many reasons for a person (or many people) to want to block an user, and not all of them are the reasons that would generate a report.

Therefore, the suggestion is: at least in gallery images, have a report button (the ones that usually have a flag), so the user can easily and directly report to admins, so they can properly ban the user, check what made them be able to make CSAM (and therefore enhance the safeguards), and so on.

To avoid excessive useless reports, other things may be applied, for example:

• Cloudfare/whatever to avoid bots.
• Quick rules on reporting (for example, "Use this only for X, and never for Y, since for Y you must simply block or whatever").
• If people keep reporting for Y instead of the proper X, you may put a selection box with X and Y, and auto-ignore everything that has only Y.
• If people still abuse the report button, you may make this button appear only for logged in users, which are easier to deal with.

Obviously, like the title says, this is only a suggestion. I don't really know the things about taking care of a site and all, and I know that Perchance already does a good job. Also, I'd like be thankful for perchance, because it's one of the only remaining sites around the web that won't get random keywords and send my IP to the cops because of my prompt "Smiling cute little girl, looking at viewer, holding a white board written, in red letters, "donate blood" in it".

Edit: I jut noticed "report button" exists on images one create themselves, but not for others. The objective is obviously for other people's images, since someone who wants to create CSAM will hardly report themselves.