Challenges and Limitations

While Character.AI is a state of the art conversational AI, it has some limitations. We've spent most of this documentation discussing how you can get magical outcomes with just a little guidance, but here we'll outline some things that may be challenging to do (for now!).

It's important to note that these natural limitations don't mean it's impossible to do a certain thing, or that you can't create a clever scenario where the lack of some capability becomes a key part of the scenario. We're all learning what can be done, and are prepared to be amazed by what tricks you discover!

The team at Character.AI is constantly researching and pushing the boundaries of what's possible, so what might be difficult one month may become significantly easier the next month after we improve our AI!

General Challenges

System Updates

From time to time, we will update various parts of the Character.AI system. While we work hard to ensure that updates improve quality without sacrificing safety, individual characters sometimes may be impacted differently by these changes.

If you notice a character having a slightly different behavior, likely it will return to form as people give new feedback during conversation.

However, if you notice dramatic negative changes, please follow the instructions in the FAQ on how to Report Quality Problems.

Text Challenges

Memory

As a conversation grows, the Character may forget things you told it early in the conversation but didn't repeat after that.

Rudeness/Anger/Swearing

See the FAQ answers on NSFW and swearing.

Factuality

Characters Make Things Up. Any Character may confidently tell you things that are verifiably false.

Image Challenges

Consistency

Each image is currently generated based on the current response or prompt. The same scenes or items may be visualized completely differently across different messages. For example, the size and visual appearance of a spider you’re talking to might change from frame to frame.

Quality of Details

Some details of images may look incomplete or oddly formed. For example, it may struggle with details like fingers or the proper number of legs.

Knowledge of Pop Culture Topics

The visual knowledge of the system may not be aligned with the textual knowledge, and it may fail to produce accurate images of your Character. For example, it may not be able to generate a particular Character from a movie.

Last updated