4 ms·
The AI predictions based on Moore's law type reasoning by Kurzweil, Moravec etc have been pretty accurate and not subject to the it's always 10 years ahead thin
by tim333 2y ago
The AI predictions based on Moore's law type reasoning by Kurzweil, Moravec etc have been pretty accurate and not subject to the it's always 10 years ahead thing.
- zero-sharp 2y agoOh ok. I thought we were talking about the article (or at least claims that are just as bold): "Deep learning works, and we will solve the remaining problems." "It is possible that we will have superintelligence in a few thousand days (!)"
- tim333 2y agoIt was more in reply to "people in the AI community have"... Which some of them have but the Moravec type stuff has been quite accurate. Technically a few thousand days covers quite a range. 20 thousand is 55 years. On the Kurzweil graph, extrapolating hardware progress from 1900 through 2000, superintelligence seems to be roughly 2035, depending on how you define things. https://www.researchgate.net/figure/Kurzweils-8-71-chart-of-exponential-growth-of-computing_fig1_333518196 https://www.researchgate.net/figure/Kurzweils-8-71-chart-of-...