AI Companies' Secret Models May Be Months Ahead, Experts Warn

Imagine using a flip phone while the company that makes it secretly has the latest smartphone. That's the situation with AI: the most powerful models may be kept internal, never released to the public. This gap means what you see isn't what they have, and it could mislead everyone about how fast AI is really advancing.

One clue is how much AI companies are spending internally on computing power for research. A graph shows spending by OpenAI researchers growing at an explosive rate—if it continued, the median researcher would spend $185 million a year on AI, and top researchers $2.3 billion. That's like a single person using the computing power of a small country. Such growth can't last, but it signals intense effort to push AI further.

Another sign: revenue for frontier AI companies is growing 3-4 times per year. If that keeps up, the industry would be larger than the entire world economy by the early 2030s. That sounds crazy, but it shows how quickly AI is becoming a economic force. More importantly, AI is starting to automate its own research and development. One analysis suggests that within a couple of years, AI could handle most of the work of building better AI—a scenario that could lead to rapid, unpredictable changes.

These 'holy shit' graphs are a wake-up call. They suggest we might be just a few years away from AI that can improve itself without human help. That could bring enormous benefits but also serious risks, like loss of control. The companies should share this data so policymakers and the public can prepare. The catch? These are projections, not guarantees—but ignoring them could leave us blindsided.