Why this CEO thinks video games make better training data than the internet

When it comes to achieving artificial general intelligence (AGI), large language models just don’t have what it takes. Models like ChatGPT and Claude are great at text, but they’re less skilled at understanding how things actually move through space and time — an essential skill for producing intelligence that generalizes. That gap, it turns out, might be filled by gaming data. That’s the bet behind General Intuition, a company that’s leveraging video games as a primary source of training data for its AI models.

The CEO of General Intuition, who has spent years researching and developing AI systems, believes that video games offer a unique advantage when it comes to training AI models. “The problem with the internet is that it’s primarily text-based,” he explains. “While text is great for learning about abstract concepts and relationships, it doesn’t provide the same level of spatial and temporal understanding that’s required for true intelligence.”

Video games, on the other hand, offer a rich source of data that’s inherently spatial and temporal. Games are designed to simulate real-world environments, complete with physical laws, cause-and-effect relationships, and complex interactions between objects. By training AI models on gaming data, General Intuition aims to give its models a more comprehensive understanding of the world — one that goes beyond mere text-based knowledge.

But why do video games make better training data than the internet? According to the CEO, it all comes down to the type of data that’s available. “The internet is full of noise and bias,” he says. “There’s so much text data out there, but most of it is irrelevant or misleading. Gaming data, on the other hand, is highly structured and curated. Games are designed to be engaging and challenging, which means that the data they generate is inherently meaningful and relevant.”

Another advantage of gaming data is that it’s often annotated with rich metadata. Games typically include detailed logs of player behavior, including actions, decisions, and outcomes. This metadata provides a wealth of information about the game world, including the relationships between objects, the consequences of different actions, and the strategies that players use to succeed. By leveraging this metadata, General Intuition can train its models to understand the underlying dynamics of the game world — and, by extension, the real world.

Of course, one of the biggest challenges of using gaming data to train AI models is the sheer volume of data that’s available. Modern games can generate terabytes of data per day, including gameplay logs, video feeds, and other types of telemetry. To make sense of this data, General Intuition has developed a range of specialized tools and techniques, including custom-built data pipelines, distributed computing architectures, and advanced machine learning algorithms.

Despite the challenges, the CEO of General Intuition is confident that gaming data holds the key to unlocking true AGI. “We’re not just talking about building better language models or computer vision systems,” he says. “We’re talking about creating intelligence that can generalize across multiple domains and tasks. That requires a deep understanding of the world, including its spatial and temporal structure — and that’s exactly what gaming data provides.”

To achieve this goal, General Intuition is focusing on a range of applications, including robotics, autonomous vehicles, and smart cities. In each of these domains, the company is using gaming data to train AI models that can understand and interact with the physical world. For example, by training models on data from racing games, General Intuition can develop AI systems that can navigate complex road networks and avoid obstacles — skills that are essential for autonomous vehicles.

The potential benefits of this approach are enormous. By leveraging gaming data to train AI models, General Intuition could help to accelerate the development of AGI, enabling a wide range of breakthroughs in fields like healthcare, finance, and education. At the same time, the company’s focus on gaming data could help to reduce the costs and risks associated with AI development, making it more accessible to a wider range of organizations and industries.

So what does the future hold for General Intuition and its innovative approach to AI development? According to the CEO, the company is already making rapid progress, with several high-profile partnerships and collaborations in the works. As the demand for AGI continues to grow, General Intuition is well-positioned to capitalize on the trend, providing a unique and powerful solution for organizations that need to develop intelligent systems that can generalize and adapt.

As the CEO puts it, “We’re not just building AI models — we’re building a new type of intelligence that can change the world. And with gaming data on our side, we’re confident that we can make it happen.”

The Benefits of Gaming Data

So why do video games make such good training data for AI models? Here are just a few of the benefits:

• Rich spatial and temporal structure: Video games are designed to simulate real-world environments, complete with physical laws, cause-and-effect relationships, and complex interactions between objects. This provides a rich source of spatial and temporal data that can help AI models understand how the world works.
• Highly structured and curated data: Games are designed to be engaging and challenging, which means that the data they generate is inherently meaningful and relevant. This reduces the risk of noise and bias in the data, making it easier to train accurate and reliable AI models.
• Detailed metadata and annotation: Games typically include detailed logs of player behavior, including actions, decisions, and outcomes. This metadata provides a wealth of information about the game world, including the relationships between objects, the consequences of different actions, and the strategies that players use to succeed.
• Scalability and volume: Modern games can generate terabytes of data per day, providing a vast and diverse range of training data for AI models. This scalability is essential for developing AGI, which requires massive amounts of data to learn and generalize.

The Challenges of Gaming Data

While gaming data offers many benefits for AI development, it also presents several challenges. Here are a few of the key issues:

• Volume and complexity: The sheer volume of gaming data can be overwhelming, requiring specialized tools and techniques to process and analyze. Additionally, the complexity of gaming data can make it difficult to extract meaningful insights and patterns.
• Noise and bias: While gaming data is generally more structured and curated than internet data, it can still contain noise and bias. For example, player behavior may be influenced by factors like demographics, experience, and motivation, which can affect the accuracy and reliability of the data.
• Domain adaptation: Gaming data is typically generated in a specific domain or environment, which can make it difficult to adapt to other domains or contexts. For example, a model trained on data from a racing game may not perform well in a game that involves combat or exploration.
• Computational resources: Processing and analyzing large volumes of gaming data requires significant computational resources, including specialized hardware and software. This can be a major challenge for organizations that lack the resources or expertise to handle large-scale data processing.

Conclusion

In conclusion, video games offer a unique and powerful source of training data for AI models, providing a rich and diverse range of spatial and temporal information that can help to develop true AGI. While there are challenges associated with using gaming data, the benefits are clear — and companies like General Intuition are already making rapid progress in this field. As the demand for AGI continues to grow, it’s likely that gaming data will play an increasingly important role in the development of intelligent systems that can generalize and adapt to new contexts and challenges.