Questworld
Living game worlds
TL;DR:
Questworld is a “living” game world where the characters, content, and environments are all autonomous and evolving. It combines many generative functions to give each player a unique experience with profound depth and immersion.
- Characters have minds of their own, content is generated from player behavior, and environments are intrinsically mutable.
- Pushes the frontier of AI in video games by vertically integrating many generative features on a foundational level.
- To start we're building a text-based RPG world from the ground up, but ultimately this can be a platform that goes beyond games.
- Cofounders David Byttow and Aaron Krane have built AAA games and startups before, while raising from investors like Founders Fund, Sequoia, Khosla, KPCB, GV, Index, etc.
- Raising a modest first round to build the technology foundation and start beta testing the first version.
Problem Statement
Today, generative AI and LLM-based products are getting a lot of attention for performing narrow tasks. For example, you can use AI products to write a passage, or generate an image, or chat with an agent. But it's still early days, and we think that many of these products are actually just pieces of larger puzzles. The future lies in combining many such features into generalized experiences, and one such example is living virtual worlds.
Living virtual worlds are a concept with potentially broad applications, but one of the most immediate is in video games. Billions of people spend trillions of hours in video game worlds, only to churn due to repetitiveness. While modern video games have tried to solve this with both user-generated and procedural content, they've only managed to add shallow breadth to their worlds, when what they need is depth…
Solution
Questworld is a “living” game world, where the characters, content, and environment are all autonomous and evolving. Characters are more than conversationalists, they have minds of their own and their actions shape the world around you. Content, storylines and quests expand continuously, unconstrained by canned scripts. And worlds persist and evolve, growing deeper with every experience. We think this will redefine immersion in games, and lead to a revolution even larger than 3D graphics.
Autonomous characters
Questworld does not have typical NPCs (non-player characters). Instead, it has autonomous characters with minds, memories, and motives of their own.
Characters are no longer defined by scripted behavior, but rather are free to act based on scope, context, and experience. Imagine entering a tavern where the inhabitants actually remember you from your last visit, just like real people. That is, except for the drunk fisherman in the corner, who always forgets and needs a reminder or two. But if you keep going back then eventually he might remember you too, and share the location of a sunken treasure. Whether you believe him is up to you, because after all, he's a drunk fisherman with a mind of his own.
Character behavior will be an ongoing process of tuning, in which we are directors giving feedback to an AI cast. Every experience of every character will be recorded, and characters will be interviewed in order to understand their thinking and adjust accordingly. The goal is that characters be able to observe, recall, and react in ways that are both scoped to their personas and compatible with the ecosystem. This fidelity, once achieved, will be a tipping point for immersion.
Expanding content
In Questworld, the content, storylines and quests continuously expand, so as in life, you know where it starts, but no one knows it will go.
Quests and storylines are generated from prompts. For example, "a mysterious wizard seeks an artifact hidden in the ruins of Elarion." With this, generative AI fleshes out the locations, characters, and events in order to form a rich story. First it adds a new section to the world, and in it places a bustling town, a shadowy forest, and the crumbling ruins of Elarion. In the forest it places a tower, and at the top of the tower it places Azerel, a reclusive wizard. Finally, it generates other key details, like the market where Azerel recruits adventurers, the overgrown path to his tower, and the labyrinthine ruins, filled with puzzles and spirits.
Once generated, quests expand based on player behavior. For example, if upon retrieving the artifact you decide to inspect it, then you might discover strange etchings. If you show the etchings to a historian in town, then you might learn that the artifact contains an ancient evil. If you still decide to give the artifact to Azerel, then he might use it to unleash a dark spell into the forest, turning wildlife into monsters that attack the town. All of this would be generated in response to specific player behavior, resulting in unique and unpredictable content with miraculous depth.
Evolving worlds
The environments of Questworld are symbiotic with the content, so environmental interactions are far from superficial.
Any interaction with the environment can become a key input in the world's generative feedback loop. For example, feeding a stray dog might result in that dog approaching you the next day. Feeding it again might result in it following you around, and training it for a few weeks might result in it sniffing out traps and treasures. Now, a seemingly ambient detail has evolved into a game-changing companion.
But such an integration of generative AI is more complex than adding ChatGPT to a SaaS app. It requires vertical integration and complex coordination on a low level. That's why we are building a new game world from the ground up, with mutability and evolution as first principles. If we are successful, then this can expand into a platform for others to create worlds as well.
Platform
As we build this living game world, we'll also build the tools for others to do so.
We don't know exactly what this will look like, but we believe that by first focusing on a compelling product, we'll quickly learn how to best enable the platform (like DOOM circa 1993). This platform will be 10x easier than Roblox, because it replaces design and scripting with prompts and curation. And it will be 100x deeper, because the outputs live and evolve autonomously from their creators.
The tools that we build will not only be powered by, but also operated by AI. We already have custom GPT “assistants” to fill out characters, quests, and environments. Soon, we'll train our own AI to use these assistants and recursively generate more content. And while today there are major gaps in logical and spatial reasoning, upcoming models like GPT4.5 and 5 will 10x these capabilities, and we'll be positioned to leverage them on arrival.
Opportunity
We are initially targeting the video game market, which is expected to grow from $227B to over $310B in the next four years. A single game world has made over $3B (Minecraft), and a platform of static user-generated worlds is the #2 grossing iOS app of all time (Roblox).
It's also possible that a platform for living simulation has applications in other industries, and the actual TAM is larger than games.
Go to market
Our goal is to build large-scale living worlds, but we're starting with a few text-based rooms. Our first version will be a very small and visually simple fantasy RPG, because our initial focus is on character behavior, content generation, and environmental interaction. Once these building blocks have been refined and integrated into a solid foundation, then we'll add graphics that rival any AAA game studio.
In Q1 of this year, v1 will be released to about 500 beta testers who are primarily avid players of RPGs. We will source them from our network and player referrals, and access will remain restricted as we develop the world. We will use this scarcity to build a large waitlist for subsequent releases.
FAQ
What will gameplay be like?
Questworld is a text-based, adaptive and interactive role-playing game where you create and take control over your character and their identity. You will run into characters that are controlled by players like you, and yet others that are controlled by AI (NPCs). Similar to any RPG, you will be able to gain experience through hunting and quests in order to increase your level, learn new skills, and find treasures in the form of weapons, gold, and other items. This will allow you to enter new areas, unlock more difficult quests, and gain notoriety within Questworld among other characters. You can either choose to go solo or join with others in your adventures, of course, the latter will make the experience more enjoyable and slightly less impossible.
Even when you're not playing, the world continues to evolve both in its landscape and inhabitants. Every day, new NPCs and characters alike will enter and exit the world in order to shape it.
What are the NPCs like?
NPCs are rich characters with their own motivations and backgrounds. They are always watching (unless they're paying attention to something else) and learning about what's going on around them. For example, a simple example of a player taking an action and asking Dolores (NPC) what happened.
This experience breathes new life into the idea of interacting with an autonomous agent, as with any simulation, NPCs experience time and space. This can and will lead to many unique experiences that we couldn't have imagined only a couple years ago.
Many NPCs simply go about their lives without any custom programming or scripting. They're simply given a backstory, motivations, and the set of actions they can take. For example, one was simply told that they are drunk and like to wander around muttering to themselves.
Why build a text-based game now given the popularity of high-budget, large-scale 3D game worlds?
- Familiarity and Technology Synergy. Text-based environments are a natural starting point due to their synergy with the current state of LLMs, which excel in text processing and generation, making them ideal for creating rich, narrative-driven environments. This also aligns well with our initial target audience and gaming community's historical familiarity with text-based RPGs like Dungeons & Dragons.
- Complexity Management. Starting with a text-based game allows for managing the complexity of AI-driven NPCs. Text interfaces are less resource-intensive and more forgiving of AI limitations than full 3D environments. It's easier to iterate and refine AI behavior in text format before tackling the additional complexities of visual representation.
- Nostalgia and Revival of Classic Gaming Formats. We can foster and build on a resurgence of interest in indie/retro gaming and classic formats, including MUDs. This project can tap into this nostalgia, offering a modern twist with advanced AI integration, appealing to both old-school gamers and new audiences.
- Innovative Storytelling. This project aims to revolutionize storytelling by allowing NPCs to have their own motivations and story arcs. This is more feasibly tested and refined in a text-based game where the focus is on narrative and interaction, rather than graphics or action.
- World Building. The tools we are building enable anyone with the desire to build and populate a world (large or small) full of life. Text-based games simplify this because one can leverage LLMs themselves to help create the rich content.
Why do games need “living” worlds?
Generative AI lowers the barrier to adding new content and evolution to the world. Sentient and complex characters dramatically increase fun and immersion. Quests and stories constantly expand and never go stale. Environments are miraculously and deeply interactive. The idea of an evolving simulated world is a simple and compelling one, but has been out of reach for games (notably that didn't have a critical mass of players).
Who is the market for living game worlds?
Initially, the market will be players of RPGs and online game worlds. As the simulation gets more realistic, the market will broaden to more casual game players like Fortnite, Roblox. Ultimately, immersive living simulations should have applications outside of gaming.
What's your go-to-market and growth strategy?
We are starting with a hand-picked group of 500-1000 beta testers as we develop the basic generative features. However, we do anticipate getting a certain amount of attention from media and WOM. We'll leverage that to build a waitlist, which could get quite large by the time we make a public release.
Is the content fully autonomous?
It's a hybrid of editorial + generative. We start with prompts written by people, to ensure basic quality. We use these prompts to generate quests and stories for the world, and then these quests and stories expand over time with more generative content.
Could the content ever be fully autonomous?
In the near term no, in the medium/long-term yes. Generative AI isn't good enough at composing quests and worlds yet, and more research is needed in this space. But everyone underestimates how rapidly it will improve. We'll be building and learning along the way.
Are we in a simulation?
Who are your competitors?
Competition will come from indie game studios, many of whom are currently leading the pack by testing ChatGPT. But their integrations are superficial and their tech/ML is not their strong suit. We will build a premiere AI team to stack the tech deck in our favor.
Why won't large studios do this soon?
They've actually used procedural generation for years to create maps, terrain, enemies and so on. However, they don't move quickly enough to refactor existing titles any time soon, and their new titles take more time than Hollywood blockbusters, so they are unlikely to be early movers.
Is this just a ChatGPT wrapper?
No, this is the vertical integration and complex coordination of multiple technologies on a very low level. The result is a multifaceted simulation that is greater than the sum of its parts.
Why can't you just add this to existing games?
Existing games can handle superficial integrations of AI, like procedural terrain, mediocre dialogue, or background music. However, in order to add deeper features like autonomous characters and evolving quests, you must build a game world from the ground up.
What are the technology challenges?
There are many challenges involved in the coordination of a multifaceted generative world. For example, visual world generation, contextual content expansion, character LLM steering and contextual memory, prompt completion costs and latency, etc.
What is your product strategy?
Our strategy is to build a game world from the ground up, so we can vertically integrate many ai features on a very low level. We will start with a text-only experience to focus on the foundational technology, and add graphics later. Unlike game studios, we will ship early and iterate quickly.
How can I help?
We're looking for people who want to be on the ground floor and are willing to roll up their sleeves and join us on this adventure. We are looking for:
- Engine and AI Programmers. We need talented coders who are interested in building a platform that enables GameMasters to build rich experiences. You should have strong software engineering skills and be skilled in Python. You will either be building core game systems that are flexible and scalable or developing and tightly integrating LLMs to create rich NPC experiences.
- GMs (GameMasters). GMs may wear many hats and be talented in a variety of areas that are relevant to the game they work on. We are looking for GMs who excel in one or more of the following roles:
- Designer. Designers are GMs that help design narratives, storylines, quest areas, monsters, NPCs, and everything else content in the game.
- Builder. Scripting behaviors for monsters, NPCs, quests, and other mechanics. These are more technical GMs that implement designs that either they come up with themselves or with other GMs
- Testers and Early adopters. We need feedback from people who love games and worlds like the one we're building. You should be ready and willing to give a lot of feedback and work closely with the GMs.
Sources
- https://www.globaldata.com/store/report/gaming-market-analysis/
- https://www.pwc.com/gx/en/industries/tmt/media/outlook/insights-and-perspectives.html
- https://www.statista.com/statistics/273829/average-game-hours-per-day-of-video-gamers-in-selected-countries/
