Pada tahun 2012 saya berpartisipasi dalam diskusi kelompok kecil yang dipimpin oleh @doctorow. Saya kebetulan melihatnya, tetapi dia melukiskan gambaran yang sering saya pikirkan selama bertahun-tahun sejak itu, dan itu telah mengubah cara saya melihat dunia.

Here's a great paper by Nobel winner Philippe Aghion (and Benjamin F. Jones and Charles I. Jones) on AI and economic growth. The key takeaway is that because of Baumol's cost disease, even if 99% of the economy is fully automated and infinitely productive, the overall growth rate will be dragged down and determined by the progress we can make in that final 1% of essential, difficult tasks. And this logic still applies *even* in a world with AGIs that can automate *every* task a human can do. In this world, the "hard to improve" tasks would no longer be human-centric ones, but physics-centric ones. The economy's growth rate stops being a function of how fast/well the AGI can "think" and starts being a function of how fast it can manipulate the physical world. Essentially, post-AGI does not necessarily mean post-scarcity: the entire cost and value of the economy becomes concentrated in the physically constrained tasks: generating energy, mining resources, manufacturing goods, transportation and so on.

Saya bersumpah setiap organisasi penelitian telah berjuang dengan: "Bagaimana kita berbagi lebih banyak WIP tanpa orang memperlakukannya sebagai final?" Suka bagaimana mengklik halaman Catatan baru @METR_Evals mengubah seluruh situs menjadi font tulisan tangan dan latar belakang kapur. Visual yang kuat berteriak "tidak serius, ini kasar".