August 17, 2021

A New Story Experience on AI Dungeon

We have some news to share about upcoming improvements to the user experience in AI Dungeon. We’ve agreed upon a new approach with OpenAI that will allow us to shift AI Dungeon’s filtering to have fewer incorrect flags and allow users more freedom in their experience. The biggest change is that instead of being blocked from playing when input triggers OpenAI’s filter, those requests will be handled by our own AI models. This will allow users to continue playing without broader filters that go beyond Latitude’s content policies.  

In the long term, this will be a much better experience for users where we will have more flexibility to define our own content policies, listen to user feedback, and make sure our filters aren’t hurting the player experience.

In the short term, we recognize that some users on the Dragon AI might experience lower performance as a result of requests that trigger OpenAI’s filters being routed to our models. We are working on a longer-term solution that will provide all users with the highest quality AI. In the meantime, we are introducing several improvements for users with Dragon access to provide our subscribers with extra value, including:

  • Early access to visual generation mode. Initially, this mode will generate basic landscapes that match your current location, but we will improve the generation quality over time. (This received almost 1000 votes on the feature upvote board!)
  • Double the current memory context (to 2048 tokens) when not on OpenAI models.
  • Auto summarization. The AI will automatically summarize your story periodically to improve memory.

These and other changes will roll out soon, starting August 23rd. We believe that this will lead to a much better experience for all users in the long run, but for anyone who feels the Dragon experience isn’t what they expected, just reach out and we will be happy to give a refund of the last month.

We welcome your feedback and will take any suggestions into careful consideration. The past several months have been a wild ride, but we’ve been working hard to make AI Dungeon the best experience we can. We won’t always get things right, but we’re trying our best and want your feedback on how we can do better. Thank you for playing AI Dungeon!

Additional Questions

What changed with the filter?

As most of you know, we’ve been testing various filters and suspension systems. We recognize that these systems have had false positives and unintended side effects that we know have frustrated many of you (trust us, we haven’t wanted those either). We regret the frustration these caused and the way we had to roll those out.

What some may not know is that, behind the scenes, we needed to implement these changes to prevent AI Dungeon from being required to use the much stricter content filter that would normally be required for use of OpenAI’s GPT-3 models. That filter often blocks even mild fantasy violence, which would obviously hurt the magic of AI Dungeon—we believe that you should be able to fight off bandits, slay dragons, and have all of the kinds of epic adventures that make AI Dungeon so fun.

While we’ve been working with OpenAI to find a joint path forward, it’s clear that we have different approaches to providing AI experiences to users safely, and that Latitude needs to be able to decide on the approach that best supports our users. Because of this, we have decided to transition to a new implementation that gives us more flexibility in how we let you experience the magic of AI.

How will the filter work with these changes?

If your story triggers OpenAI’s filters it will be handled by our own models rather than resulting in account suspension. We will still use and continue to work on our own filters to enforce our content policies, but they will be much more carefully targeted, ensuring that you aren’t unfairly penalized for incorrectly flagged content.

We will also continue to work on our own model improvements as well as 3rd party models that our users can access. This approach means that users who stay within OpenAI’s content policies can still use OpenAI’s models, but other users won’t be blocked by stricter filters than Latitude feels are required.