Things I read and liked in August
Self-respect, solar, SB 1047, space progress, surprise without surcease
Yuxi Liu gives us a fascinating deep dive on Marvin Minsky, the Perceptron Controversy, and the history of scaling in AI.
The cutting edge is cool, but the dull edge of technology is underrated: rearranging cheaper, older technology in new ways. Mary Hui gives the examples of the Game Boy, intermittent windshield wipers, and chiplets.
They say that countries win more medals when they host the Olympics. But do they? And if so, why? asks Dynomight.
Steve Newman looks at the heated debate over California’s AI regulation bill, SB 1047. Among those who (thoughtfully) disagree about the bill, he argues, a key crux is how much AI capabilities will advance. We need to gather better information about AI progress and dangers—and that “is a job for policy, not just voluntary commitments and privately funded research”.
There's a place for everyone: Adam Mastroianni on finding your niche.
Norm MacDonald tweet from 2018 (i.e. from a dying man): “At times, the joy that life attacks me with is unbearable and leads to gasping hysterical laughter. I find myself completely out of control and wonder how could life could surprise me again and again and again, so completely. How could a man be a cynic? It is a sin.”
Solar power has gotten way cheaper in the last 15 years. Clean Energy Review: How do we make solar even cheaper?
T. S. Eliot, The Cocktail Party (1948)
Half of the harm that is done in this world
Is due to people who want to feel important.
They don't mean to do harm—but the harm does not interest them
Or they do not see it, or they justify it
Because they are absorbed in the endless struggle
To think well of themselves.
(source)
Given the amazing things that AI can create, it’s a bit pedantic to get hung up on the question of whether they are creative, argues Rosie Campbell: “AI systems may soon be capable of discovering cures for cancer, creating novel pathogens, and even contributing to their own improvement…In the face of systems with such profound impacts, quibbling over definitions seems a little...uncreative.”
Étienne Fortier-Dubois meditates on the evil and beauty of the Venus Flytrap. Reminiscent of Annie Dillard’s Pilgrim at Tinker Creek.
The Grey Matter is irked by something that has also irked me: studies that “measure the energy consumption of using AI to do things and compare it against… nothing”.
Jeff Fong on the puzzle of San Francisco’s empty downtown offices. Vacancy rates as high as 30%, still way down post-COVID and remote work—why can’t rents just lower until the occupancy rate is higher?
Richard Ludlow shares his favorite meditation resources.
Dean Ball wonders just how interpretable, in principle, complex AI systems can be (maybe not very). Related: Chris Olah on the "dark matter" of interpretability.
Joan Didion, “On Self-Respect” (1961):
To have that sense of one's intrinsic worth which, for better or for worse, constitutes self-respect, is potentially to have everything: the ability to discriminate, to love and to remain indifferent. To lack it is to be locked within oneself, paradoxically incapable of either love or indifference. If we do not respect ourselves, we are on the one hand forced to despise those who have so few resources as to consort with us, so little perception as to remain blind to our fatal weaknesses. On the other, we are peculiarly in thrall to everyone we see, curiously determined to live out—since our self-image is untenable—their false notions of us. We flatter ourselves by thinking this compulsion to please others an attractive trait: a gift for imaginative empathy, evidence of our willingness to give…. At the mercy of those we can not but hold in contempt, we play roles doomed to failure before they are begun, each defeat generating fresh despair at the necessity of divining and meeting the next demand made upon us.
Does ChatGPT have a mind? A philosophical analysis of LLM mentality by Simon Goldstein and Ben Levinstein.
Sarah Constantin reviews the state of AI for biology: Although “it makes sense to be excited about the field”, she says, “AI optimists who aren’t familiar with biotech are often wildly miscalibrated about what AI tools can do even in the best case scenario.”
Space progress is too slow! And expensive. But we can do better, says Quade MacDonald.
Thanks, love these posts!
🙏