Ethan Mollick

Three findings from a trial use of a GPT-4 tutor for Harvard's intro CS course: 1) Students found it useful (88% found it helpful) & used it a lot 2) The AI made mistakes, even with RAG (81% accurate) 3) It was pretty cheap Lots of potential but R&D needed

The O*net papers are great, but they don’t tell us how jobs will change.

This is such an interesting thread, and a lightning tour of what advanced manufacturing tools look like today, from industrial robots to lasers. (As a side note, this is also very old-school, now mostly lost, Twitter, where experts put together long threads on narrow topics)

Three hints to the future of AI in this paper: 1) It is close to outperforming the top humans at geometry (superhuman performance is possible) 2) It combines a language model & other AI approaches (hints of the value of tools) 3) It trained on AI-generated data (big for training)

I replied directly to the post. Its a great paper, and it really asks a different question than our paper (or other ones looking at performance impacts of AI). The AI isn’t doing, here, instead it is advising.

I wrote about this (excellent) paper here, in the section on mentoring. This is an example of AI giving advice, rather than doing the work, which is why high performers benefit so much. They ask better questions & are better able to execute.

I just found out that one of the more thought-provoking modern science fiction books to explore the nature of sentience is available for free via Creative Commons. As a warning, Blindsight is not particularly happy or uplifting, but it is fascinating

The first outbreak of the Plague, the 541 Plague of Justinian, was thought to have effectively ended the Roman Empire & killed over 1/3 of Europe. But new historical work hasn’t found strong evidence for large-scale disease. (But the debate continues)

There is a lot to pay attention to in this paper: 1) It is very easy to jailbreak AIs to overcome guardrails by just using the persuasion techniques that work on humans! 2) They show a wide range of techniques 3) The page about the paper is just amazing

The three existing recordings of Fitzgerald (none of them reading his own work):

A 2021 paper finds teaching with AI-generated historical figures can aid learning: [link] This is now easy. Based on his voice, here is simulated F. Scott Fitzgerald reading the end of The Great Gatsby. Is it too creepy? (Yet... cartoon Einsteins are everywhere)

I think a key point of the book is the opposite of what you are saying: we underestimate the changes in quality of life for the vast majority in the US over a very short period of time