OGWiseman Rounds Up!
Some interesting things I found on the internet!
Another busy fortnight on the A.I. and beyond science and tech beat. I never feel like I’m reading as much as I used to or as I’d like to, but I reach the end up a couple of weeks, check all the stuff I’ve emailed to myself throughout the period for possible newsletter inclusion, and somehow there are always a bunch. Here’s some favorites:
This is the most interesting and well-written thing I have read this month: “trees are harlequins, words are harlequins”. A strange name, but a clearly-explained and user-friendly history of language models, in all their utility and contingencym leading up to an extended discussion of current state-of-the-art. Long, but really does a good job of communicating the complexity and subjectivity of many outstanding problems in the field. An excellent sort of starting place if you’re just starting to really learn about this.
In the above essay I found this one linked: A blog post from “Janus”, a twitter personality I’ve followed for a long time, about the complexities of using synthetic data to train AI. An excellent read from a very unique and unusual thinker who is as in the trenches with A.I. as it is possible to get.
“What does Elon Musk really think?” has become on of the enduring cultural and political mysteries of our time, and of course I don’t have a definitive answer, but given that he is the richest man in history and also was on track to be the Thomas Edison of the 21st century, it matters. One thing I do know is that his answers about A.I. in this recent interview:
Just don’t make an ounce of sense in the context of his recent behavior. He says that A.I. is like a “1000 foot tsunami” that about to come and wash all our other concerns away, which is a plausible position, but if you believed that, why would you go have a huge fight and spend time, energy, and political capital on the U.S. budget deficit? In a strong A.I. scenario either our economy will grow so fast the current debt won’t matter, or the entire concept of “debt” will be outmoded by our much larger problems. Why didn’t he try to do something on A.I. policy? Such a frustrating waste of a brilliant and ambitious mind!
This will be the subject of a future story of mine. I mean, of course this was going to happen. Of course! A.I. is a mirror. All technologies are, in a way, but A.I. is *really* a mirror, my goodness.
Riley Goodside is someone I only know as a Twitter account, but he does really interesting testing with A.I. models, using recursion and unclear referents to construct logic puzzles at the frontier of current models’ reasoning abilities. One example:
I happened upon the list by Jacob Trefethen of his favorite science writing from the last two years, which sent me down a number of rabbit holes, and I read at least a dozen worthwhile and interesting pieces. A few gems:
A Primer on why Microbiome research is hard. As someone who has struggled with stomach issues and seen a bunch of doctors who seemed clueless and frankly useless, it was nice to have my feelings validated by this.
Speaking of stomachs, this dude gave himself dysyntery on purpose for charitable research. There has got to be an easier way! It was a funny read, though.
Universal Antivenin (snake poison antidote) may be found in blood of man who let snake bite him over 200 times. More a story about fate and chance than anything else. The man’s brother died shortly before 9/11, and that combined with the tragedy of the day made him start self-harming with snake bites, and now that may turn out to be the most important and valuable thing he ever did with his life. You just never know!
A list of interesting questions about biotech. Some of these are too technical for me, but I found them provocative and informative.
A compelling meditation on causality in biological systems, a gloss on the current state of medical research, and a subtextual argument about the practical importance of semantics. This post is like catnip for me!
How virologists lost the gain-of-function debate. Ask A.I. first if you’re not familiar with gain-of-function. The importance of this rather esoteric bioethics question is highly, highly underrated at this point. There’s a nonzero chance that this becomes the most important thing in the world one day.
A compelling history of the centuries-long effort to treat cholera, finally ending in “oral rehydration solution”, a simple misture of water, salt, sugar, and some zinc, that has saved the lives of 70 million children since it was introduced in the 1970s. Lesson: It is maddeningly difficult to discover things and keep them discovered. Metalesson: Discovering technologies sooner has incredible benefits, and the technologies we have not currently discovered should often be treated like emergencies!
A useful primer on the concept of “p(doom)”. If that term doesn’t mean anything to you, definitely go read this.
END
Thanks as always for spending part of your Sunday with me! I hope everybody has a great week, and I will be back next Sunday with another original story!