Datapoint from a late-twenties Ph.D. student in AI:
During my formative years studying pure math, I learned that the hard work of _caring to understand_ is rewarded---for example, by your internal sense of beauty and by the enthusiasm of other mathematicians.
For my Ph.D. I decided to try bringing the mathematical toolbox to the field of artificial intelligence, which seemed to be facing some interesting problems. Overall, the message I've received loud and clear is that understanding is _not rewarded_ in this field, and actually discouraged.
The work of an outwardly successful machine learning scientist is to make incremental improvements to well-established methods and package the results in easily marketable papers. Learning about and using previous work is rewarded to the extent that it improves marketability (for example, by improving Greek alphabet or \displaymath density) but is discouraged if standing on the shoulders of giants in this way renders the paper incomprehensible to a goldfish with a knowledge of high school algebra.
I think it's clear that our current paradigm on deep learning is struggling to make scientific progress. The hard work of reshaping a paradigm must involve looking back and _caring_ about the perspectives that have been developed over the last 60-odd years. Learning other points of view is hard work, but it's what you're supposed to do as a scientist! Unfortunately, it's hard to overstate how little people care about anything beyond their diminutive research niche. (I had one colleague who said their work hasn't been intellectually demanding ever since LLMs got good at writing pytorch code and others who seem to feel that LLMs are "experts on math.") Every day I look at my own research group and the boatload of "top conference papers" and feel sad and angry that we call this science.
It's certainly poetic that LLMs---the emblem of people who don't care---is backed by a "research community" that, on average, doesn't know what it means to care about science.
During my formative years studying pure math, I learned that the hard work of _caring to understand_ is rewarded---for example, by your internal sense of beauty and by the enthusiasm of other mathematicians.
For my Ph.D. I decided to try bringing the mathematical toolbox to the field of artificial intelligence, which seemed to be facing some interesting problems. Overall, the message I've received loud and clear is that understanding is _not rewarded_ in this field, and actually discouraged.
The work of an outwardly successful machine learning scientist is to make incremental improvements to well-established methods and package the results in easily marketable papers. Learning about and using previous work is rewarded to the extent that it improves marketability (for example, by improving Greek alphabet or \displaymath density) but is discouraged if standing on the shoulders of giants in this way renders the paper incomprehensible to a goldfish with a knowledge of high school algebra.
I think it's clear that our current paradigm on deep learning is struggling to make scientific progress. The hard work of reshaping a paradigm must involve looking back and _caring_ about the perspectives that have been developed over the last 60-odd years. Learning other points of view is hard work, but it's what you're supposed to do as a scientist! Unfortunately, it's hard to overstate how little people care about anything beyond their diminutive research niche. (I had one colleague who said their work hasn't been intellectually demanding ever since LLMs got good at writing pytorch code and others who seem to feel that LLMs are "experts on math.") Every day I look at my own research group and the boatload of "top conference papers" and feel sad and angry that we call this science.
It's certainly poetic that LLMs---the emblem of people who don't care---is backed by a "research community" that, on average, doesn't know what it means to care about science.