Beyond Chatbots: A New Approach to Game Discovery
In a landscape often dominated by AI-generated content and chatbot interfaces, Eli, a developer using the handle Death12th on Reddit, has launched GameCombiner. This project offers a refreshingly direct approach to game discovery. It eschews large language models (LLMs) for a more deterministic, vector-based system that mathematically combines game attributes to suggest new titles. The core idea is to move beyond simple genre tags and explore deeper relationships between games, surfacing recommendations that might otherwise remain hidden.
GameCombiner operates on a dataset of 146,288 games. Each game is represented as a 1024-dimensional vector. These vectors are not derived from game titles, but rather from a rich combination of user-generated tags, genres, and store descriptions. This granular data allows for a more nuanced understanding of what makes a game unique and how it relates to others in the vast gaming catalog.
The Math Behind the Mix
The process of combining games in GameCombiner is rooted in linear algebra. When a user selects two games to combine, the system retrieves their respective 1024-dimensional vectors. It then calculates the mathematical midpoint between these two vectors. This midpoint represents a conceptual space that lies between the two chosen games, embodying a blend of their core attributes. The system then searches its entire database for the game whose vector is closest to this calculated midpoint. The result is a real game from the catalog, not a synthesized description or a novel concept.
This method ensures that every combination yields a tangible result from the existing game library. Crucially, it means that combining the same pair of games will always produce the identical outcome. There is no randomness or AI interpretation involved in the selection process. This deterministic nature is a key differentiator from many contemporary discovery tools that might leverage LLMs for more subjective or varied suggestions. Eli emphasizes that no LLM is involved in choosing the answer, and nothing is generated. The output is always a real catalog entry, and the same pair always returns the same result.
Why This Matters: Coherence and Predictability
The implications of this approach are significant for users seeking reliable and understandable recommendations. Unlike black-box recommendation engines or LLM-based suggestions that can sometimes feel arbitrary, GameCombiner offers a transparent mechanism. The developer, Eli, actively invites feedback on the coherence of the results and offers to explain the reasoning behind specific combinations. This interactive element transforms game discovery from a passive experience into an exploratory dialogue.
Consider the current state of game discovery. Platforms often rely on collaborative filtering (people who liked X also liked Y) or content-based filtering (games with similar tags). While effective to a degree, these methods can struggle to uncover less obvious connections. GameCombiner's vector-space model allows for a more fluid exploration of the game catalog. For instance, combining a complex strategy game known for its intricate resource management with a fast-paced arcade shooter might yield a surprising but relevant indie title that balances both elements in its gameplay loop. The system is not limited by predefined categories but by the actual mathematical relationships between game features.
A Toy, Not a Tool?
Eli describes GameCombiner as a "small game discovery toy." This framing perhaps undersells its potential. While it may not be a full-fledged recommendation engine with user profiles and historical data, its unique methodology offers a valuable perspective. For developers, it provides a novel way to understand game relationships and potentially identify gaps or opportunities in the market. For players, it offers a fun, interactive way to stumble upon games they might never have found through conventional means.
The decision to exclude game titles from the vector generation is particularly interesting. Titles can be misleading or generic. By focusing on genres, tags, and descriptions, GameCombiner aims to capture the essence of a game's mechanics and player experience. This is akin to understanding a book by its plot synopsis and thematic elements rather than just its title. The resulting vectors represent a distilled form of game identity, allowing for comparisons that transcend superficial naming conventions.
Future Directions and User Engagement
The project is currently free to use, and Eli is actively seeking user input. Comments on the coherence of the results are particularly valuable. By sharing combination results, users can engage directly with the developer, gaining insight into the system's decision-making process. This open feedback loop is crucial for refining the underlying data and potentially the vectorization techniques themselves.
What nobody has addressed yet is how this vector-space approach could scale to other media. Could a similar system combine films based on plot summaries and critical reviews, or music based on lyrical themes and instrumental composition? The underlying mathematical principles are transferable, suggesting that GameCombiner might be the first glimpse into a broader paradigm for content discovery that prioritizes intrinsic attributes over superficial metadata.
The platform is accessible at https://gamecombiner.com. It represents a thoughtful departure from the current trend of AI-driven content generation, offering a more grounded and predictable, yet equally intriguing, method for exploring the vast universe of video games.