Phoenix Beta Release & Product Updates

We have been working on one of our most transformational AI Dungeon updates ever: Phoenix. Today begins a beta testing period to provide all our players the opportunity to give us feedback before we move forward with a broader release.

Let’s start with the best part: AI Dungeon has been overhauled to the core with a completely rebuilt technology stack and a redesigned front-end interface.

There are many awesome parts of this release and this blog post will walk through 4 key areas we want to provide more detail on:

  • Redesign
  • Improved AI
  • AI Safety Settings
  • Beta Environment

All of these updates will increase stability, reduce bugs across various platforms, and empower our team to more easily add new elements that upgrade and enhance the game experience over time.

Redesign

When AI Dungeon was first developed, the team was primarily composed of developers and lacked user experience designers. They did a great job making AI Dungeon into a successful game, but not having designers on the team resulted in AI Dungeon feeling clunky and unintuitive at times. Since then, our team has added additional talent who brought important skill sets, including UX design, front-end development, mobile, UX, etc. These awesome team members have allowed us to take the game to a new level, not only designing to meet player needs but also setting a stronger foundation for more successful future releases. This brings us to the redesign announced today.

Each page and feature has been rethought, revised, refined, and refreshed, but you’ll find that, while everything may look different, it’s still instantly familiar. The main game elements and essential play features are still there, but everything around them has been elevated and evolved to be more elegant and intuitive.

To highlight a few main user themes we focused on:

  • Interactions—Every tap or click is part of the interaction of the site and defines how users interact with content, other players, and the AI.
  • Information Architecture—How the site is organized and how content is displayed on each page. Players need to be able to find what they are looking for in a way that makes sense.
  • Focus—We aim for every page to have a clear primary action, purpose, and function, where important information is always prioritized.
  • Hierarchy—The design needs to ensure that the most important elements stand out through the use of size, space, color, and other design elements.
  • Typography—As a text-based game, the use of type is incredibly important. We have considered readable fonts that pair well, and can be used with different weights and sizes to help with information hierarchy.
  • Accessibility—It’s important to us that everyone can use the site. We designed with accessibility in mind, architecting the site in a way that allows everyone to enjoy.
  • Performance—No one likes a slow app. We’ve spent time making sure you don’t have to waste yours.
  • Intuitive play—We’ve made the first round of many improvements to ensure a better first-time user experience for players just discovering AI Dungeon.

A few in-game changes we want to share with you:

Site-wide navigation

With fewer, clearer navigation choices, players will more easily be able to discover the key actions they can take. Daily rewards, notification center, and profile buttons are now grouped together and readily accessible.

Home page

We’ve redesigned the banner section at the top of the home page. The banner section will continue to provide helpful information about updates, events and tips, but now in an even more elegant way. The biggest change to the home page is our new content carousels. We’re introducing several new categories to help players discover more great content. The page will be refreshed regularly so you always have fresh content to explore.

Our new Creators Program is also reflected in the carousel. Each quarter, we identify our top and emerging creators and this group will now have their work featured prominently on the home page.

Content cards

All of our content cards on the site now appear with images. Players have the opportunity to pick an image that best represents their published Adventure, Scenario, or World from a new library of AI-generated images. We know you’d like even more, so we’re exploring ways to enable players to add images that aren’t part of our image library.

“New” button

We replaced the “Play” button with “New” in the upper-right corner on desktop to make it easier for players to quickly start an Adventure and also provide a more efficient way for creators to access Scenario or World templates.

On mobile, you can quickly start an Adventure through the big plus sign at the bottom of your screen. The sliding pop-up lets you quickly choose between an Adventure, Scenario, World, or Multiplayer game to create.

Profile

Many of you will be pleased to see the return of a sorely missed feature: the ability to view a list of your followers, following, and friends. Also under the profile menu is the User Settings page. This User Settings page has been simplified to include basic global actions, like changing your username, email, password, linked accounts, or membership. All other game settings are only available in-game from the play screen now.

Game Screen

Our team has been spending significant time and resources doing more thorough user research, and one area that consistently gives players trouble is the gameplay screen. In particular, players often don’t understand several important actions. The Do/Say/Story/See actions are hidden, and it isn’t clear you can change your action type. Many players also don’t realize they can simply continue the story without typing or taking a turn.

In the game, we now have “Take a Turn”, “Continue”, “Retry”, and “Erase” as our primary action buttons. “Take a Turn” is a new addition that immediately creates a more intuitive role-play experience within AI Dungeon. Selecting this actually opens a brand new way to use our essential Input Modes: Do, Say, Story, and See.

We’ve also made sure that typing your next action is only one click or tap away, just like the legacy version of AI Dungeon. In AI Dungeon today, the text box is always visible, but it still requires a tap to begin typing. With Phoenix, as soon as you tap “Take a Turn”, the cursor is populated into the text box so you’re ready to go immediately.

The old paper airplane icon has been retired in favor of a “Continue” button, which is a much more straightforward way to display and explain that function for new users.

The “Continue” and “Retry” actions now have equal importance alongside the “Take a Turn” action. This is significant because it will help players familiarize themselves with these important actions more quickly.

“Erase” is now included in an effort to separate undo/redo from story actions and include broader user actions, so instead of “undo” meaning “erase”, “erase” means “erase”, which is then an action that can be undone.

In-game settings are still accessible through the gear icon in the upper right corner of the play screen. They’ve been simplified into three categories: Gameplay, AI Models, and Appearance.

You can access the Memory, Author’s Note, and World Info context tools under the Gameplay tab. Under the AI Models tab, you can change the language model you’re playing with, choose a different image generation model, or opt into or out of our “Improve the AI” program. The Appearance tab lets you customize certain aspects of your AI Dungeon experience.

The new game screen features our flame icon in the upper left corner instead of a back arrow. The title of your current Adventure is listed next to the icon so you always know what story you’re playing. Selecting the flame icon brings up links to your profile and resources. A small “Now Playing” card appears below those with the title of your current Adventure, a tiny thumbnail picture, and the option to “Exit Adventure” back to the Home screen.

Learn More About the Redesign

If you want to see more specific details, Devin, our Director of Experience, walked through a few pages as part of the announcement livestream. You can also see Phoenix for yourself by accessing our new beta site at.

AI Safety Settings

As part of the Phoenix release, you’ll see an overhauled AI Safety Setting feature. We previously offered “Safe Mode” within AI Dungeon, and over the last few months, we’ve seen increased user feedback that this feature wasn’t providing the expected amount of control players wanted over their experiences. Players sometimes encountered uncomfortable situations with the AI, even when Safe Mode was turned on.

As we considered this feedback and reexamined this feature, we realized that determining the safety of AI-generated content and protecting our players from potentially harmful material is a very difficult problem to solve, and frankly it’s one we haven’t done well on in the past.

We started looking for a technology partner who specializes in content safety and could provide a better solution than we have. We identified HiveAI as that solution, and have implemented their AI-based scoring technology within AI Dungeon to give players better control over their experience and help them feel safe when interacting with the AI. Depending on the player’s safety settings, as they engage with the AI their model outputs will be sent to Hive for a safety score. If the AI-generated output doesn’t meet the threshold set by the player, we will either provide a different generation or let the user know there was an issue.

As with any partnership, we've taken time to evaluate Hive’s product and terms of service to ensure our players' privacy and security is protected. Hive is well-established in the machine learning field and their product is utilized by many top brands and companies. None of the data we send to Hive will be stored, looked at, or used in any way to train their models. All API requests are encrypted, processed by Hive’s internal servers, and then immediately deleted. Their use of our players’ content is only to generate a safety score during gameplay. We’re excited that Hive is becoming an integral part of AI Dungeon and helping us protect our players.

Our new AI Safety Setting has three levels: Safe, Moderate, and Mature. These correspond roughly with familiar movie ratings. The Safe setting is similar to PG, Moderate is closer to PG-13, and Mature is R, which is the same as our current mature 18+ setting. These levels let players dictate what kind of content the AI can generate, and we can change and adjust them as we receive feedback.

Along with this, we also improved the safety settings for Discover and Search. We’ve intentionally set up this setting to operate independently of the AI Safety Settings, since we know some players like to create their own NSFW content, but prefer not to see other people’s NSFW content when browsing. You’ll see a new search-specific filter in Discover that will simply display or not display NSFW-tagged results. The AI Safety Setting and Search Filter both default to Safe, and players can change those themselves. Additionally, any published worlds, scenarios, or adventures, will be sent to Hive and receive a safety score, to help prevent mistagged content from showing up in the wrong places.

Ultimately, our updated AI Safety Setting and Safe Search Filter are just more ways players get to customize and control their individual AI Dungeon experience. They provide additional safety controls for those who want them, while still offering flexibility for those who don’t.

Improved AI

We’re experiencing a renaissance in AI right now, with new models and innovations hitting the market faster than ever. We’ve been capitalizing on these innovations, and are excited to announce two new models to be added to AI Dungeon.

New Dragon

Dragon has traditionally been perceived by our players as our best model. In the last few months, Wyvern has eclipsed Dragon in many ways, taking the spot of players’ most preferred premium model. Now, the balance is shifting back to Dragon as it evolves with our new update.

The new Dragon will be based on the Jurassic-2 Grande model provided by AI21 Studios. This model is powerful, but we’ve taken it even further with new enhancements we’ve recently developed. The new Dragon model has quickly become a favorite with our Alpha testers. Some of our longtime players consider the GPT-3 based version of Dragon we offered a few years ago as the gold standard of AI models. The new Dragon is so good that Alpha testers are reporting that it’s exhibiting the same quality and storytelling capabilities they remember from the old Dragon.

AI21’s new Jurrassic-2 models have eclipsed their old models in every way. Because of that, they’ve announced they will be ending support of their old Jurassic-1 models. As a result, we’ll be retiring Legacy Dragon and Wyvern. However, given the success of our tests with the new Dragon, we’re confident these outgoing models will not be missed.

The new Dragon offers several key advantages over any other version of Dragon we’ve offered before. The new Dragon is fast—responses come in at nearly half the time as Legacy Dragon. This has always been one of the disadvantages of Legacy Dragon, and Alpha testers have already shared how much the improved speed has improved their play experience. The new Dragon will also support more options and possibilities for training and improving the model over time. And probably most exciting the new Dragon will support larger memory. Adventure will get 1x memory, Hero will get 2x, and Legend will get 4x. This means the AI can consider more of your story when generating the next output, ensuring key details and historical story events are not missed.

The new Dragon will replace Wyvern as the new default model for our AI Dungeon subscribers. It will be available in both the Phoenix beta, and the production version of AI Dungeon.

ChatGPT Beta

We’re also excited to introduce ChatGPT through Microsoft Azure as a new beta model in AI Dungeon. Players have shared with us how impressed they are with the quality of ChatGPT, and we received numerous requests to add it as a playable model in AI Dungeon. In fact. our March player survey showed that over 70% of players who responded were excited about the idea of ChatGPT + AI Dungeon.

The partnership with Microsoft Azure is a new one for us, so we are taking it slowly, and have sought out player feedback at every step to ensure that this is what’s best for AI Dungeon.

For context, Microsoft is the only company besides OpenAI to have access to the ChatGPT model. We chose to use Microsoft Azure for a few specific reasons, the most important being their privacy and security reputation. This is crucial to our players’ experiences in AI Dungeon, as well as to us.

Some players may wonder if Azure sends data to OpenAI and they do NOT. Microsoft hosts the OpenAI models within their own Azure infrastructure, and all customer data sent to Azure remains within the Azure service. This means that data processed by Azure is never sent to OpenAI. Microsoft runs ChatGPT on their own servers with strong security and privacy protections they’ve developed as part of the Azure stack.

We utilized our entire AI evaluation tool kit to determine whether we should offer ChatGPT from Azure long-term, and early testers have compared it to being as good as or better than old Dragon, so we’re excited to start offering this model alongside our others. During this testing period, Adventurers can use ChatGPT with the standard memory offering, Hero-tier subscribers will have access to 2X memory, and Legend-tier to 4X memory.

ChatGPT has performed extremely well for adventures and has the potential to greatly improve areas where other models have struggled in the past, like memory, repetition, and creativity. However, we will continue to gather more feedback, answer questions, and address concerns before moving the model to production full-time.

Using ChatGPT is entirely opt-in for premium users only. With the new version of Dragon, our premium players will have two fantastic models to choose from for their adventures. A new Griffin model is also being developed based on the improvements we implemented in the new Dragon.

Beta Environment

Starting today, we are launching a brand new Beta Environment of AI Dungeon for Phoenix. We previously offered players a chance to try new Beta features by enabling beta features as a setting on their account. This new, completely separate environment lets players explore new features, and then have the option to go back to the stable version if desired.

The previous approach worked well in the early days of AI Dungeon when new features were being added almost daily. It allowed developers to move quickly. Moving to a standalone beta environment will allow new features and changes to be further isolated from the stable version of AI Dungeon, ensuring that new developments don’t break the core experience.

Going forward, new features and major updates will first be released to this Beta Environment. In beta, changes will be thoroughly tested, reviewed by players, and validated before being added to the stable version of AI Dungeon.

The redesign we’ve announced today will only be available in the beta environment during the testing period. After we do more testing, migrate all legacy features, gather feedback from players, and validate everything is working as expected, we'll deploy this Phoenix update to production and officially retire the legacy version of AI Dungeon.

Not all legacy features will be available at the start of the Beta testing period. Features that are still in progress for inclusion in Phoenix in the coming weeks include:

  • World Info
  • Multiplayer
  • Native Mobile Apps
  • Scripting
  • Image generation (See Mode)
  • Train the AI

However, the data flows seamlessly between the Beta and Production version of AI Dungeon, so if there’s a feature that’s missing, you can still use it in the Legacy version.

You can access Beta via: beta.aidungeon.com using same login credentials as you use for AI Dungeon.

Coming Soon

Phoenix addresses much of the remaining player feedback we discussed in our October 2022 blog post, “Next Steps on AI Dungeon,”  by bringing even more stability, better AI quality, and refreshed designs to AI Dungeon. Of course, maintaining and further improving AI Dungeon is an ongoing process that we look forward to working on with our next releases. We also have plenty of ideas to continue growing AI Dungeon and will put out a new blog post detailing those plans when we’re ready to talk about them.

Thank you to our entire player base, alpha testers and Discord contributors–your knowledge and feedback are invaluable to us and vital to providing a better experience for all players. We are extremely grateful for your dedication and commitment to AI Dungeon!