3 ms·

I generally agree, although an important aspect of thinking in terms of "trajectories" is recognizing when a particular trajectory might end up at a dead end.

I generally agree, although an important aspect of thinking in terms of "trajectories" is recognizing when a particular trajectory might end up at a dead end. One perspective on the weaknesses of current LLMs is that it's just where the things are today and they can still provide value even while the technology improves. But another perspective is that the persistence of these weaknesses indicates something more fundamentally broken with the whole approach that means it's not really the path towards "real" AI, even if you can finesse it into doing useful things in certain applications. There's also an important nuance differentiating rejection of a general technological endpoint (e.g. AGI or Level 5 self-driving cars) with a particular technological approach to achieving those goals (e.g. current LLM design or Tesla's autopilot). As you said, "never" is a long time and it takes a lot of unwarranted confidence to say we will never be able to achieve goals like AGI or Level 5 self-driving. But it seems a lot more reasonable to argue Tesla or OpenAI (and everyone else doing essentially the same thing as OpenAI) are fundamentally on the wrong track to achieving those goals without significantly changing their approach. I agree that none of that really warrants dismissive cynicism of new technology, but being curious and future-looking also requires being willing to say when you think something is a bad approach even if it's not totally useless. Among other reasons, our ability to explore new technology is not limitless, and hype for a flawed technology isn't just annoying but may be sucking all the oxygen out of the room not leaving any for a potentially better alternative. Part of me wants to be optimistic about LLMs, but another part of me thinks about how much energy (human and compute) has gone into this thing that does not seem to be providing a corresponding amount of value.

wenc 2y ago
I appreciate this thoughtful comment. You are absolutely right that the trajectories, if taken linearly, might hit a dead end. I should clarify that when I mentioned "trajectories" I don't mean unpunctuated ones. I am myself not convinced that LLMs -- despite their value to me today -- will eventually lead to AGI as a matter of course, nor the type of techniques used in autopilot will lead to L5 autonomy. And you're right that they are consuming a lot of our resources, which could well be better invested in a possibly better alternative. I subscribe to Thomas Kuhn's [1] idea of scientific progress happening in "paradigms" rather than through a linear accumulation of knowledge. For instance, the path to LLMs itself was not linear, but through a series of new paradigms disrupting older ones. Early natural language processing was more rule-based (paradigm), then it became more statistical (paradigm), and then LLMs supplanted the old paradigms through transformers (paradigm) which made it scale to large swaths of data. I believe there is still significant runway left for LLMs, but I expect another paradigm must supplant it to get closer to AGI. (Yann Lecun said that he doesn't believe LLMs will lead to AGI). Does that mean the current exuberant high investments in LLMs are misplaced? Possibly, but in Kuhn's philosophy, typically what happens is a paradigm will be milked for as much as it can be, until it reaches a crisis/anomaly when it doesn't work anymore, at which point another paradigm will supplant it. At present, we are seeing how far we can push LLMs, and LLMs as they are have value even today, so it's not a bad approach per se even though it will hit its limits at some point. Perhaps what is more important are the second-order effects: the investments we are seeing in GPUs (essentially we are betting on linear algebra) might unlock the kind of commodity computational power the next paradigm needs to disrupt the current one. I see parallels between this and investments in NASA resulting in many technologies that we take for granted today, and military spend in California producing the technology base that enabled Silicon Valley today. Of course, these are just speculations and I have no more evidence that this is happening with LLMs than anyone else. I appreciate your point however and it is always good to step back and ask, non-cynically, whether we are headed down a good path. [1] https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Revolutions https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Re...

wenc 2y ago

I appreciate this thoughtful comment. You are absolutely right that the trajectories, if taken linearly, might hit a dead end. I should clarify that when I mentioned "trajectories" I don't mean unpunctuated ones. I am myself not convinced that LLMs -- despite their value to me today -- will eventually lead to AGI as a matter of course, nor the type of techniques used in autopilot will lead to L5 autonomy. And you're right that they are consuming a lot of our resources, which could well be better invested in a possibly better alternative. I subscribe to Thomas Kuhn's [1] idea of scientific progress happening in "paradigms" rather than through a linear accumulation of knowledge. For instance, the path to LLMs itself was not linear, but through a series of new paradigms disrupting older ones. Early natural language processing was more rule-based (paradigm), then it became more statistical (paradigm), and then LLMs supplanted the old paradigms through transformers (paradigm) which made it scale to large swaths of data. I believe there is still significant runway left for LLMs, but I expect another paradigm must supplant it to get closer to AGI. (Yann Lecun said that he doesn't believe LLMs will lead to AGI). Does that mean the current exuberant high investments in LLMs are misplaced? Possibly, but in Kuhn's philosophy, typically what happens is a paradigm will be milked for as much as it can be, until it reaches a crisis/anomaly when it doesn't work anymore, at which point another paradigm will supplant it. At present, we are seeing how far we can push LLMs, and LLMs as they are have value even today, so it's not a bad approach per se even though it will hit its limits at some point. Perhaps what is more important are the second-order effects: the investments we are seeing in GPUs (essentially we are betting on linear algebra) might unlock the kind of commodity computational power the next paradigm needs to disrupt the current one. I see parallels between this and investments in NASA resulting in many technologies that we take for granted today, and military spend in California producing the technology base that enabled Silicon Valley today. Of course, these are just speculations and I have no more evidence that this is happening with LLMs than anyone else. I appreciate your point however and it is always good to step back and ask, non-cynically, whether we are headed down a good path. [1] https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Revolutions https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Re...