Reactions from ranked influencers
5 posts
New tech does not change the world overnight. In 1806, the steam engine was over 90 years old and Watt's improved engine over 35 years old. But there were no practical steam-powered boats, trains or flying machines; and most factories were not yet automated. Diffusion takes time. Even if it is 10x faster today, if ChatGPT ~= the Watt engine of AI, then we are only at about 1806 right now. The steamboats and locomotives of AI are coming soon. And then the equivalent of the internal combustion engine and the electric generator…
I think a reasonably fair question to ask AI ppl these days is something like, “If the models are so powerful, why does the world seem so normal?” https://twitter.com/cliffordsosin/status/2078594661359194500
BTW, in 1806, all of the following was true: - The engines then in widespread use were not good enough to power locomotives. They were too heavy; they needed to use high-pressure steam to increase the power-to-weight ratio - A working prototype of a locomotive, with a high-pressure engine, had already been demonstrated, but hadn't yet been developed into a practical railroad system. Getting to a workable passenger train, more than 20 years later, would require a lot of engineering on the railroad and the locomotive itself, not just the engine - The engines that did exist were not yet in use for everything they could be useful for, e.g., powering factories; diffusion was still ongoing - Taking full advantage even of the capabilities of the engine at the time required more than just people waking up to the opportunity; it required lots of new machines to be invented that would be powered by the engine - None of the above was evidence against the enormous potential of steam power in particular and even less so against industrialization in general By analogy, it is probably the case today that: - not all applications of even today's AI capabilities have been implemented yet - fully realizing the power of even today's models/agents will require engineering tooling and processes around them - models/agents will continue to get better and will be able to do even more things in the future So the lack of bigger impact today might say something about the power of today's models, but not as much as you might think, and it doesn't say much about the potential for AI in general.
BTW, in 1806, all of the following was true: - The engines then in widespread use were not good enough to power locomotives. They were too heavy; they needed to use high-pressure steam to increase the power-to-weight ratio - A working prototype of a locomotive, with a high-pressure engine, had already been demonstrated, but hadn't yet been developed into a practical railroad system. Getting to a workable passenger train, more than 20 years later, would require a lot of engineering on the railroad and the locomotive itself, not just the engine - The engines that did exist were not yet in use for everything they could be useful for, e.g., powering factories; diffusion was still ongoing - Taking full advantage even of the capabilities of the engine at the time required more than just people waking up to the opportunity; it required lots of new machines to be invented that would be powered by the engine - None of the above was evidence against the enormous potential of steam power in particular and even less so against industrialization in general By analogy, it is probably the case today that: - not all applications of even today's AI capabilities have been implemented yet - fully realizing the power of even today's models/agents will require engineering tooling and processes around them - models/agents will continue to get better and will be able to do even more things in the future So the lack of bigger impact today might say something about the power of today's models, but not as much as you might think, and it doesn't say much about the potential for AI in general.
@stevesi True. This is why I suggested applying a ~10x multiplier on the timeline
@jasoncrawford One thing that has changed has been network effects. The more something requires or benefits from connectivity the faster the adoption. Baseline technology is itself quite an accelerator.