A lot of very smart people, both within NLP research, and here, are in complete denial about what the proliferation of high quality LLMs means for their jobs and earning potential.
The only thing which makes me less sad is that I'm pretty sure Moravecs paradox is actually not all that real, but is more due to the relative lack of engineering interest put on solving continuous control problems. Apparently reinforcement learning on transformers works now (RLHF in ChatGPT). This implies that we should see high effective continuous control models very soon. Robots are coming for physical labor, it'll just take a bit longer.
Shit man, when I was last in South Korea, I felt like I was living in the future. They had many "24/7" drink cafe's where it's literally just a robot arm that makes the drink for you for a few dollars.
It is painfully ironic to knowledge workers that they are destroying their own earning potential, but physical labor is not safe. Nothing is safe.
The only thing which makes me less sad is that I'm pretty sure Moravecs paradox is actually not all that real, but is more due to the relative lack of engineering interest put on solving continuous control problems. Apparently reinforcement learning on transformers works now (RLHF in ChatGPT). This implies that we should see high effective continuous control models very soon. Robots are coming for physical labor, it'll just take a bit longer.
Shit man, when I was last in South Korea, I felt like I was living in the future. They had many "24/7" drink cafe's where it's literally just a robot arm that makes the drink for you for a few dollars.
It is painfully ironic to knowledge workers that they are destroying their own earning potential, but physical labor is not safe. Nothing is safe.