The Overhyped World of Large Language Models
Ah, large language models (LLMs) like ChatGPT and Claude. They're the darlings of the tech world, boasting user numbers that make social media platforms blush. But let's not kid ourselves—these models are far from the omnipotent AI overlords some would have you believe. In fact, they're pretty clueless when it comes to understanding how objects move through space and time. You know, the kind of basic understanding you'd expect from a toddler.
The Limitations of LLMs
According to some brave souls willing to speak the truth, "large language models just don’t have what it takes." These models are great at spitting out text that sounds human-like, but when it comes to grasping the physical world? Not so much. This glaring gap in their capabilities is a significant roadblock on the path to achieving Artificial General Intelligence (AGI).
Enter General Intuition: The Startup Betting on Gaming Data
While the rest of the world is busy feeding LLMs more internet data, General Intuition is taking a different route. They're betting that video games, with their rich, interactive environments, could provide the spatial and temporal understanding that LLMs sorely lack. "That gap, it turns out, might be filled by gaming data," they claim.
Why Video Games?
Video games are not just about mindless entertainment. They're complex systems where objects move, interact, and change over time. This makes them a goldmine for training AI to understand the real world. Unlike the static, text-based data from the internet, gaming data offers dynamic scenarios that could teach AI how to navigate and interpret the physical world.
The Market Implications
The implications of this shift are significant. For one, the video game industry could see a surge in demand as a source of training data. Meanwhile, the AI development market, particularly in regions like Russia, might pivot towards integrating gaming data into their models.