06/11/2025
Here's a friendly (there's an animated cat professor presenting, LOL) — though admittedly somewhat complex (e.g., references such as AI Takeoff, a function of 10^36 FLOPS, and the CES Production Function) — reality-based presentation of an analysis of the future evolution of Generative Artificial Intelligence. Two key takeaways from this analysis are: On the one hand, even if we experience a scenario with no breakthrough jumps in the progress of AI evolution, AI could easily race past Human capabilities in just a few years. Also, even if reaching Human capabilities is a more difficult achievement for AI than we expect and AI progress is generally slower than we expect, it's likely that world-changing AI systems are merely a few decades away; and it's very hard to find a realistic set of inputs for this analysis that doesn't get us to AI's ability to perform all Human cognitive tasks by 2060. (Importantly, this does depend on the assumption that it's possible to build superhuman AI with the current LLM-driven paradigm. Notwithstanding, of course, new paradigms may also be developed.)
Here's a friendly (there's an animated cat professor presenting, LOL) — though admittedly somewhat complex (e.g., references such as AI Takeoff, a function of 10^36 FLOPS, and the CES Production Function) — reality-based presentation of an analysis of the future evolution of Generative A...