"The distance between econometrics theory and applied work can shrink rapidly." This is one of the positive outcomes that is essentially available and under leveraged right now.
I appreciate your thoughtful treatment of this topic. Most economists I've read seem to agree with your contention that the skills that LLMs have not depreciated become more valuable.
And I agree with your suggestion that the increased pace enabled by LLMs needn't mean that we are deluged with slop. Being able to execute the mechanics faster could lead to better papers because researchers will have more time to hone their ideas and analysis.
In any endeavor where there are positive returns to iteration, speed can be used in the service of quality. Early in my career I learned this about software development. It seems intuitive that researchers who learn to use these tools will demonstrate this as well.
Economists understand that a lawyer who is a the best typistist in the world should still delegate typing since the opportunity costs of foregone legal work while typing outweigh the cost of hiring a secretary. If science were a rational enterprise, then applying the insights of comparitive advantage to scientific labour in the age of AI should cause us to fundamentally reallocate the division of labour between people and machines, resulting in more overall scientific output. Human scientists have a comparitive advantage in *judgement*, implying we should allocate human capital on activities like peer-review rather than writing. Alas the incentives of academia are oriented around a model in which scientific progress is measured in terms of the numbers of papers published in "prestigious" journals, even though writing is a only a small part of the scientific process. Meanwhile peer-review is not even measured and is therefore not incentivised. It is only by changing these incentives and metrics that we will be able to fully exploit the gains of AI in science. I have written about this here: https://sphelps.substack.com/p/science-has-a-slop-problem-ai-didnt
"The distance between econometrics theory and applied work can shrink rapidly." This is one of the positive outcomes that is essentially available and under leveraged right now.
I appreciate your thoughtful treatment of this topic. Most economists I've read seem to agree with your contention that the skills that LLMs have not depreciated become more valuable.
And I agree with your suggestion that the increased pace enabled by LLMs needn't mean that we are deluged with slop. Being able to execute the mechanics faster could lead to better papers because researchers will have more time to hone their ideas and analysis.
In any endeavor where there are positive returns to iteration, speed can be used in the service of quality. Early in my career I learned this about software development. It seems intuitive that researchers who learn to use these tools will demonstrate this as well.
Economists understand that a lawyer who is a the best typistist in the world should still delegate typing since the opportunity costs of foregone legal work while typing outweigh the cost of hiring a secretary. If science were a rational enterprise, then applying the insights of comparitive advantage to scientific labour in the age of AI should cause us to fundamentally reallocate the division of labour between people and machines, resulting in more overall scientific output. Human scientists have a comparitive advantage in *judgement*, implying we should allocate human capital on activities like peer-review rather than writing. Alas the incentives of academia are oriented around a model in which scientific progress is measured in terms of the numbers of papers published in "prestigious" journals, even though writing is a only a small part of the scientific process. Meanwhile peer-review is not even measured and is therefore not incentivised. It is only by changing these incentives and metrics that we will be able to fully exploit the gains of AI in science. I have written about this here: https://sphelps.substack.com/p/science-has-a-slop-problem-ai-didnt
Perhaps the title of professor should be changed to “curator” in this new era of research and graduate training?