4 ms·

As an NLP researcher, I've come across no convincing research indicating LLMs are "the possible digital twin of human intelligence." Claims certainly get made,

As an NLP researcher, I've come across no convincing research indicating LLMs are "the possible digital twin of human intelligence." Claims certainly get made, but they are subsequently shot down by future research. This has happened for discussions of emergence [1], reasoning [2], and other human attributes people have ascribed to LLMs. [1]: https://neurips.cc/virtual/2023/poster/72117 https://neurips.cc/virtual/2023/poster/72117 [2]: https://m-cacm.acm.org/blogs/blog-cacm/276268-can-llms-really-reason-and-plan/fulltext https://m-cacm.acm.org/blogs/blog-cacm/276268-can-llms-reall...

happytiger 3y ago
I think you’re discussing the logical extension of our current AI efforts, and I’m discussing the fully realized science fiction reality that could take decades to perfect or that would be driven by breakthroughs beyond the forecast of existing systems. That’s the difference between seeing it as a technology vs a larger directional shift in human development. Put more directly we have been experiencing the age of miniaturization, and now we are entering the age of digital intelligences. Very often early technology looks toy-like and lacks substantive real world application (despite the hype). It’s only later that deliver substantive applications and real-world change. This generation of AI is following the Gartner hype cycle perfectly thus far: https://www.gartner.com/en/articles/what-s-new-in-artificial-intelligence-from-the-2023-gartner-hype-cycle https://www.gartner.com/en/articles/what-s-new-in-artificial... That said Gartner tends to be part of the hype cycle these days, but the pattern of new technology holds true. So in short I would expect anything like a digital twin to emerge from post trough applications not the current generation, and I don't mean to imply that the current technology will create such a system but just that science fiction and futurists generally forecast or even predict this technology will generate a digital twin (in the form of an AGI) from this civilization-level line of technology exploration. All that said, I don’t think many of us technologists were thinking we would be anywhere near this level of AI technology in 2018. In five or six years we have gone what people thought would take 10-20. So, as always, linear extension predictive and mental models of technology advancement once again fails to predict the true course of technology progress, which seems to be much more dynamic than people generally consider when talking about the future.

happytiger 3y ago

I think you’re discussing the logical extension of our current AI efforts, and I’m discussing the fully realized science fiction reality that could take decades to perfect or that would be driven by breakthroughs beyond the forecast of existing systems. That’s the difference between seeing it as a technology vs a larger directional shift in human development. Put more directly we have been experiencing the age of miniaturization, and now we are entering the age of digital intelligences. Very often early technology looks toy-like and lacks substantive real world application (despite the hype). It’s only later that deliver substantive applications and real-world change. This generation of AI is following the Gartner hype cycle perfectly thus far: https://www.gartner.com/en/articles/what-s-new-in-artificial-intelligence-from-the-2023-gartner-hype-cycle https://www.gartner.com/en/articles/what-s-new-in-artificial... That said Gartner tends to be part of the hype cycle these days, but the pattern of new technology holds true. So in short I would expect anything like a digital twin to emerge from post trough applications not the current generation, and I don't mean to imply that the current technology will create such a system but just that science fiction and futurists generally forecast or even predict this technology will generate a digital twin (in the form of an AGI) from this civilization-level line of technology exploration. All that said, I don’t think many of us technologists were thinking we would be anywhere near this level of AI technology in 2018. In five or six years we have gone what people thought would take 10-20. So, as always, linear extension predictive and mental models of technology advancement once again fails to predict the true course of technology progress, which seems to be much more dynamic than people generally consider when talking about the future.