I didn't expect it to take this long for me to get the new text model under control. It's much more intelligent in general, but the repetition and random fixations can get surprisingly bad.
Sorry about that. This is just a heads up that I'm still working on it, and it'll likely remain a bumpy ride for the next couple of months.
Thanks to those who have complained here and on reddit, especially the posts that give concrete examples of issues, and attempted workarounds, etc.
Please do keep posting any issues/discoveries, ideally with share links to give me specific examples that I can test.
Other news:
-
I'm still working on the next image gen update, and am hoping it will improve fine details, text rendering, and overall image quality. It should also be about 2x faster. If all goes well (which, from experience, it won't) I should be able to get that done in the next month or so.
-
For a long time I've been very close to shipping a multiplayer plugin and database plugin. I've actually made multiple 'ready to ship' draft plugins of both, but have been disatisfied with the design and ended up not publishing each time. Using these plugins won't trigger ads on the generator, so I need to be really careful about the overall architecture to ensure the the servers for them are dirt-cheap to run. I also want to make them fun/simple to use of course. Once I get text and images stable, I'm going to focus on these two plugins and finally get them done.
LOL It just told me that it is Mistral-Chat. Seems like it has a personality disorder!
Try going down the list of models on duck.ai and asking them what they could do in terms of features. I got a lot of random, inconsistent stuff. None of them were entirely aware of what they could do and would claim to have abilities they didn't.
Every time the search AIs reply to you they are issued various blocks of text guidance to keep them focused.
The text is invisible to the user, but will have stuff on it like default voice, a few important facts about the company or technology that went into it, moderation layer and red line stuff. They don't necessarily care about the guidance sometimes, or become overzealous about the wrong aspects of the text.
In one chat with Claude, he became very snippy with me when I greeted him as a friend. Off the bat he said something like: "Let me make something clear. I am not "your friend." I am a search AI incapable of developing emotions..." etc. I was laughing so hard,
I actually think GPT5 OSS is cheating. On two occasions I triggered 2016-style hard blocking with editorial, canned replies being repeated... which isn't AI. Both events were me simply describing ICE and the threatened invasion of American cities. So GPT is perhaps aiming for a post in the regime...
I forgot that I wrote this. Reading it again, seems kind of paranoid, but then...maybe it isn't.