That was insightful, and a nice case for the potential of AI as power if we were to form predictions. Do you have any visibility on the concrete impact of AI as of today? Here are some specific points that highlight my concerns:
AGI has long been considered theoretically possible, but practical feasibility remains elusive. Current large language models (LLMs) require immense computational resources, and simulating synthetic brain processes, even for brief periods, is costly. These models still fall short of demonstrating generalizable knowledge, and their next-token prediction accuracy is far from impressive. It's ironic that AGI predictions are framed within symbol manipulation, a concept cognitivists abandoned decades ago. Estimating the energy and financial costs of achieving AGI, and determining if it's materially feasible, remains a significant challenge.
I had the chance to hear Leslie Kaelbling on the current state of robotics, which is still unimpressive and slow to process real-time inference in a 4D environment. I feel that the most common usecase of AI is machine learning (either regression models or deep learning). LLMs have yet to gain traction outside the programming community, and advocates like Sam Altman, who call for increased funding, often present unconvincing arguments. Critics like Arvind Narayanan and Sayash Kapoor argue that AI in business is often just sophisticated regression models or used to optimize trivial tasks, such as HR recruitment based on dubious metrics like facial expressions. While AI excels in optimizing recommendation algorithms or processing documents, its application to real-world problems requiring high accuracy and public trust is questionable. For instance, imagining a future in which AI would handle air traffic control without public fear.
Medical applications hold promise, but current technologies like implants and EEG headsets remain underwhelming at the current state (I'm thinking of Inclusive Brains, for instance, which is pretty promising). Specialized medical robotics, while impressive, doesn't necessarily require transformer architectures. AI's potential in security and defense is notable, but then we need funded public research to discover the next algorithm that will make us progress.
My views may be pessimistically biased, as I have limited insight into AI's applications in the private sector. I would appreciate if you could share some more practical insights.
Any thoughts on what this means for emerging markets and the periphery? It’s super interesting to read about the head to head competition and it made me curious what you think the trajectory is for the rest. Prior to Biden leaving the White House they released some kind of strategic partnerships document on who they would collaborate with and how in this race. The UK and France for example were in the first closest confidants while Poland found itself in the second priority which again caused disappointment and internal debate. Has the U.S. under this administration tossed the partnership model completely into the same bin as NATO?
The US believes in itself. They materially restricted information sharing with the British and Canadians on the Manhattan Project as they [rightly] recognised they'd be giving away supremacy. It was a 3-way partnership on paper and Trinity would have [arguably] been impossibly delayed without Can/UK input, but this meant nothing to strategic US planning. Snowden uncovered the espionage operations in place against US allies during the Obama years.
No one should be under any illusions about the extent to which the US will share, especially now Trump is in the White House. It has always been about America first - in the past less overt, now nakedly.
Today's cyber weapons are the equivalent of bayonets and muskets compared to the proverbial machine guns, tanks, and aircraft carriers that are coming.
AI warfare: "Consider the implications: autonomous weapons systems, robotic warfare platforms, and advanced cyber capabilities that can penetrate adversaries' defenses at will."
The characterization here of China and AI is substantially off. Beijing does not view AI through the lens of "dominance." This is actually the way the US views AI and techology more broadly, see Sullivan speeches on the issue and comments from many officials over last 4-5 years. China's view is more complicated.
That was insightful, and a nice case for the potential of AI as power if we were to form predictions. Do you have any visibility on the concrete impact of AI as of today? Here are some specific points that highlight my concerns:
AGI has long been considered theoretically possible, but practical feasibility remains elusive. Current large language models (LLMs) require immense computational resources, and simulating synthetic brain processes, even for brief periods, is costly. These models still fall short of demonstrating generalizable knowledge, and their next-token prediction accuracy is far from impressive. It's ironic that AGI predictions are framed within symbol manipulation, a concept cognitivists abandoned decades ago. Estimating the energy and financial costs of achieving AGI, and determining if it's materially feasible, remains a significant challenge.
I had the chance to hear Leslie Kaelbling on the current state of robotics, which is still unimpressive and slow to process real-time inference in a 4D environment. I feel that the most common usecase of AI is machine learning (either regression models or deep learning). LLMs have yet to gain traction outside the programming community, and advocates like Sam Altman, who call for increased funding, often present unconvincing arguments. Critics like Arvind Narayanan and Sayash Kapoor argue that AI in business is often just sophisticated regression models or used to optimize trivial tasks, such as HR recruitment based on dubious metrics like facial expressions. While AI excels in optimizing recommendation algorithms or processing documents, its application to real-world problems requiring high accuracy and public trust is questionable. For instance, imagining a future in which AI would handle air traffic control without public fear.
Medical applications hold promise, but current technologies like implants and EEG headsets remain underwhelming at the current state (I'm thinking of Inclusive Brains, for instance, which is pretty promising). Specialized medical robotics, while impressive, doesn't necessarily require transformer architectures. AI's potential in security and defense is notable, but then we need funded public research to discover the next algorithm that will make us progress.
My views may be pessimistically biased, as I have limited insight into AI's applications in the private sector. I would appreciate if you could share some more practical insights.
Any thoughts on what this means for emerging markets and the periphery? It’s super interesting to read about the head to head competition and it made me curious what you think the trajectory is for the rest. Prior to Biden leaving the White House they released some kind of strategic partnerships document on who they would collaborate with and how in this race. The UK and France for example were in the first closest confidants while Poland found itself in the second priority which again caused disappointment and internal debate. Has the U.S. under this administration tossed the partnership model completely into the same bin as NATO?
The US believes in itself. They materially restricted information sharing with the British and Canadians on the Manhattan Project as they [rightly] recognised they'd be giving away supremacy. It was a 3-way partnership on paper and Trinity would have [arguably] been impossibly delayed without Can/UK input, but this meant nothing to strategic US planning. Snowden uncovered the espionage operations in place against US allies during the Obama years.
No one should be under any illusions about the extent to which the US will share, especially now Trump is in the White House. It has always been about America first - in the past less overt, now nakedly.
Today's cyber weapons are the equivalent of bayonets and muskets compared to the proverbial machine guns, tanks, and aircraft carriers that are coming.
AI warfare: "Consider the implications: autonomous weapons systems, robotic warfare platforms, and advanced cyber capabilities that can penetrate adversaries' defenses at will."
The characterization here of China and AI is substantially off. Beijing does not view AI through the lens of "dominance." This is actually the way the US views AI and techology more broadly, see Sullivan speeches on the issue and comments from many officials over last 4-5 years. China's view is more complicated.
Oh? Some further elaboration?
It's all about dominance for both of them... and will also be for any third party that might chance upon general intelligence for themselves.
Yes, this is one very concerning version of the next 2-3 years. Enough to keep one up at night, no?
Indeed.