4 ms·

This is a common fallacy that I see often on HN. People are used to technologies that have a certain set of capabilities, where people find new ways to apply t

This is a common fallacy that I see often on HN. People are used to technologies that have a certain set of capabilities, where people find new ways to apply this technology over time. For example smartphones basically do the same kind of things they did 10 years ago, but people have found many more industries to apply them in (by writing apps). AI is fundamentally different because the capabilities of the base technology are growing rapidly, at the same time as people are finding more applications for it. So when you look into the future, you don’t just have to forecast increased applications of AI, you also have to forecast broader capabilities of the base technology. It is no use to take GPT-4’s capabilities and cast those into the future.

JimtheCoder 3y ago
"It is no use to take GPT-4’s capabilities and cast those into the future." This is true, but we cannot say "ChatGPT is huge now, and in 5 years we will have the tech to replace humans" when we have no real evidence that that is the case either...
godelski 3y ago
AI is different because it can act intelligently, getting the nuance that humans are able to. That is, it can do everything a human can do. But we're nowhere near that yet.[0] Past technology didn't directly replace human workers, but rather changed the way things were done. The difference is a self-checkout vs a janitor robot. The former is a paradigm shift, displacing the worker by shifting how the job is done and who is performing the tasks. The latter is a direct replacement of the human worker, drop in. Of course current AI will get better, but this idea has been around for decades, and like François says, people have assumed many of these problems would be solved far sooner than they have been. The thread is worth a read. We are talking about the opinions of a high ranking AI/ML researcher who extensively writes about the nature of intelligence and current limitations. Don't be fooled by fancy demos. [0] https://twitter.com/fchollet/status/1683200588782379008 https://twitter.com/fchollet/status/1683200588782379008 [1] https://fchollet.com/ https://fchollet.com/

JimtheCoder 3y ago

"It is no use to take GPT-4’s capabilities and cast those into the future." This is true, but we cannot say "ChatGPT is huge now, and in 5 years we will have the tech to replace humans" when we have no real evidence that that is the case either...

godelski 3y ago

AI is different because it can act intelligently, getting the nuance that humans are able to. That is, it can do everything a human can do. But we're nowhere near that yet.[0] Past technology didn't directly replace human workers, but rather changed the way things were done. The difference is a self-checkout vs a janitor robot. The former is a paradigm shift, displacing the worker by shifting how the job is done and who is performing the tasks. The latter is a direct replacement of the human worker, drop in. Of course current AI will get better, but this idea has been around for decades, and like François says, people have assumed many of these problems would be solved far sooner than they have been. The thread is worth a read. We are talking about the opinions of a high ranking AI/ML researcher who extensively writes about the nature of intelligence and current limitations. Don't be fooled by fancy demos. [0] https://twitter.com/fchollet/status/1683200588782379008 https://twitter.com/fchollet/status/1683200588782379008 [1] https://fchollet.com/ https://fchollet.com/