What a week/month/year it’s been in science and tech! There are so many interesting developments, most especially in A.I. but also in biotech and other areas. As the inimitable Tyler Cowen has put it, it seems like The Great Stagnation is coming to an end. And not a moment too soon.
FIRST, as fast as A.I. has been moving over the last 12-24 months, it actually seems to be speeding up. This week, Chat GPT released its API (API = Application Programming Interface, code which allows other programmers to easily integrate something into their own apps and programs). There have already been OVER 100 NEW TOOLS RELEASED! Again, that is 100 in a week.
Some of them are pretty specific applications (e.g. a program that takes photos and turns them into anime scenes), but many are complex, multilayered, and have broad use categories. One example is Trellis, which is an A.I.-powered tutor and book-reading companion. I don’t have access to the Beta as yet, but it’s not hard to ponder the implications of this kind of thing at scale. The internet has already revolutionized how many people learn things—this will do it again and should end up much more user-friendly than (for example) Youtube Tutorials.
This story is building momentum, not going away. GPT-4 (an entirely new and more powerful model, not just apps build off the old model) will appear sometime this year, and that’s all just from one company! (OpenAI) Meta, Google, Apple, they’re all building their own models with different approaches, not to mention governments of course. Again: We are in the beginning.
SECOND, here’s the weirdest thing I read this week. This paragraph:
(This is from the Dynomite substack, which I love because it’s technical enough to be insightful but not technical enough to lose me (most of the time). Please do keep in mind that the author has repeatedly said that this is highly speculative and may not be true.)
What this is referring to is the strange problem of how to collect useful *training data* for A.I. creation. Basically, the error rate (and underlying usefulness) of an A.I. is determined by some combination of the number of parameters (sort of like individual logical rules) and the amount of “tokens” (individual pieces) of training data created by humans those parameters are exposed to in order to “tune” them (eliminate errors).
It’s turning out that training data may be the more important of the two factors, and it’s turning out that there are hard limits to the amount of usable data that can be “scraped” "(downloaded and used) from all possible sources, most importantly digital books and the internet.
The problem with the internet data is that a large amount of it is garbage: It’s 4chan rants, it’s nonsense, it’s not the kind of thing you want an A.I. to be trained on, not for political reasons but because if you let the A.I. train on that data it will make stupid mistakes that cause problems.
So there’s this open question of how much of the internet’s data can be scraped in a usable form. And it turns out that the power of future A.I.s may well be determined by the answer to this question. And since the power of future A.I.s may well determine the future course of humanity itself, a really crucial, existential question for humanity may turn out to be “How much of what people write on the internet is counterproductive and stupid?” Yikes!
THIRD, on the biotech front, Weight Loss Drugs have OFFICIALLY ARRIVED. After literally decades of mediocre drug results, during which time surgery was the best and most reliable option for the treatment of obesity, we’ve arrived at a breakthrough point.
Ozempic is the name of the latest and greatest drug. It’s officially a diabetes treatment, but it’s being used off-label for weight loss so much that the drug is in critically short supply and the supplies are ramping up production (and also making an absolute killing).
For now, the still requires injections. It also requires ongoing use, unlike the surgical options, some of which are quite non-invasive and safe. Thus it’s fair to say that at this point, there’s not as much difference in quality or outcome of treatment between meds and surgery as we’d like to see. However, it will keep getting better. The breakthrough here is they’ve found the right mix of chemicals to reduce appetite. Everything else is a delivery mechanism.
A note of caution: I’m seeing this pitched in some corners of the internet/media as a “eat anything you want and never get fat” drug, and it’s just not that. It reduces appetite, but the mechanism of weight loss is still eating less, like it’s always been. The drugs make that easier (so do the surgeries), but there’s no world where you stuff your face with an entire carrot cake a day (as in my fantasy) and don’t gain a pound.
Still, this is a tremendously exciting moment in biotech generally. It’s not the most natural thought, but the Covid-19 Pandemic really did inspire an enormous leap forward in several key technology areas, the downstream results of which are just starting to arrive. Now if only we could reform the FDA more generally we could really get somewhere.
FOURTH, because this newsletter is nominally about science/sci fi literature (though I proudly go afield when the mood strikes), I’d like to recommend a book: ANALOGIA, by George Dyson. I chanced across mention of this book in a random Twitter thread, and I’m so glad I read it. It’s unclassifiable and weird and meandering, but I was riveted from beginning to end.
The book is (according to its subtitle) about “The emergence of technology beyond programmable control”, and it does get there, you sort of see what he’s getting at, but it gets there by way of a deep history of The American (and Canadian) West, really interesting information about Native American history, behavioral psychology, and countless other digressions that describe the emergence of things that we might not think of as “technologies” at all. It’s also just very engagingly written.
It all adds up to my favorite book I’ve read so far this year, and anyone with a passing interest in any of the above topics would be well-advised to pick up a copy. It’s not long, but it is dense with useful information and incisive thoughts.
That’s my three things for this week! Hope everyone has a great one, and I’ll be back next week with another original story for your enjoyment.