Favorites June 28th 2023
Metro Areas have gotten much bigger over this period, but their central cities have barely changed in size in 70 years, because we stopped allowing people to build infill housing.
Here’s the full article, it’s quite long. Some other highlights:
Providing feedback on the AI prompts is sort of “Like taking a standardized test on hallucinogens.”
Millions already doing this AI feedback work, but there’s potential for billions to be.
This description of Reinforcement Learning with Human Feedback: “ChatGPT seems so human because it was trained by an AI that was mimicking humans who were rating an AI that was mimicking humans who were pretending to be a better version of an AI that was trained on human writing."
The linked short post by Tyler Cowen was very interesting, it highlighted how being able to talk about science requires more than being able to cite individual important papers. You need an understanding of how scientists communicate by publishing papers arguing against earlier papers. Papers that have been 'disproven' are generally never retracted, even if the original author eventually concedes they were wrong. He points to a paper by Emily Oster whose conclusions were later shown to be wrong based on further study (here’s more info on that)
• Thread about our recent detection of gravity waves.