What this reminds me of, is the phenomenon in the history of philosophy, where someone thinks they have figured out the system of the world, on which successors will build. But instead what happens is that people recognize a new theme that the innovator has introduced, and build their own rival systems incorporating that new theme.
For example, Kant (responding to Humean skepticism) built his system of transcendental idealism, which was supposed to be a new foundation for philosophy in general. Instead, it inaugurated the era of “German Idealism”, which included Hegel’s absolute idealism, whatever Schelling and Fichte were up to, and even Schopenhauer’s pessimism (which in turn was a source of Nietzsche’s optimism).
Another example would be the different directions that psychoanalysis took after Freud; and I’m sure there are many other examples… I should note that in addition to the rebellious intellectual offspring, there were people who built on Kant and Freud, and who called themselves (neo)Kantians and Freudians.
The closest thing to an important technical successor to Eliezer that I can think of, is Paul Christiano, co-inventor of RLHF, a central alignment technique behind the birth of ChatGPT. Many other people must have found their way to AI safety because of his works, and specific ideas of his have currency (e.g. Jan Leike, formerly of OpenAI superalignment, now at Anthropic, seems to be inspired by Coherent Extrapolated Volition). He is surely a godfather of AI safety, just as Hinton, Bengio, and LeCun were dubbed godfathers of deep learning. But the field itself is not dominated by his particular visions.
What this reminds me of, is the phenomenon in the history of philosophy, where someone thinks they have figured out the system of the world, on which successors will build. But instead what happens is that people recognize a new theme that the innovator has introduced, and build their own rival systems incorporating that new theme.
For example, Kant (responding to Humean skepticism) built his system of transcendental idealism, which was supposed to be a new foundation for philosophy in general. Instead, it inaugurated the era of “German Idealism”, which included Hegel’s absolute idealism, whatever Schelling and Fichte were up to, and even Schopenhauer’s pessimism (which in turn was a source of Nietzsche’s optimism).
Another example would be the different directions that psychoanalysis took after Freud; and I’m sure there are many other examples… I should note that in addition to the rebellious intellectual offspring, there were people who built on Kant and Freud, and who called themselves (neo)Kantians and Freudians.
The closest thing to an important technical successor to Eliezer that I can think of, is Paul Christiano, co-inventor of RLHF, a central alignment technique behind the birth of ChatGPT. Many other people must have found their way to AI safety because of his works, and specific ideas of his have currency (e.g. Jan Leike, formerly of OpenAI superalignment, now at Anthropic, seems to be inspired by Coherent Extrapolated Volition). He is surely a godfather of AI safety, just as Hinton, Bengio, and LeCun were dubbed godfathers of deep learning. But the field itself is not dominated by his particular visions.
What about Nate soares?
What are his most important original ideas?