fancy math, becoming a bat & contrastive chain of thought prompting
CC#75 - AI and the (Danish) Labour Market, AI and CO2 Emissions & AI and Behavioural Economics
Hey there and welcome to ✨ CuratedCuriosity - a bi-weekly newsletter delivering inspiration from all over the internet to the notoriously curious.
Things I Enjoyed Reading.
🧮 Fancy math doesn't make simple math stop being true
These days I am spending quite some time hanging out with economists. In these circles, often the lingua used to describe the effect of some X on some Y, is very abstract and its easy to get caught up in talking about statistical specificities of model A vs model B. In doing so, it seems we often forget the connection to the real world, i.e. what assumptions are we making by assuming model A? This article is a friendly & entertaining reminder to economists that all the regression magic isn’t that fancy after all while intuitively explaining the concept of instrumental variables (and it’s pitfalls) to anyone who is not yet familiar with it.
I claimed this 0.443% number was biased, because acceptors are different from the general population. This sampling bias is not some abstract possibility. We know it matters, because at the end of the trial, refusers had less colorectal cancer then controls, even though neither did colonoscopies. (…)
After that article came out, I was contacted by a few economists. They said something like this:
That calculation is what we call an instrumental variables method. Because of fancy math reasons, instrumental variables methods are unbiased. So 0.443% is unbiased. Yay!
This confused me. I’d previously seen a post making this argument, but I didn’t see the point. After all, It’s the same calculation, and I trusted my argument for bias. (…)
I mean, was my argument for bias wrong? I asked everyone who contacted me what my error was, but I could never get a clear answer—the response was always to return to instrumental variables and how awesome they are. I heard lots about potential outcomes and monotonicity and latent treatment effects and two-stage least squares, but never anything about where my poor little logic went wrong.
I’m sure instrumental variables are great! (Did I mention that one of the authors of that paper won a Nobel prize for inventing instrumental variables?) But in this particular case, they produce the same number as my grug-brained logic, via the same calculation.
🧐 Tetragrammaton with Rick Rubin - Tyler Cowen [ 🎧]
I think this is the third podcast interview with Tyler Cowen that I have listened to and I am surprised how many new interesting insights I still got out of it. While there is definitely views of Tyler that I disagree with, I think his approach to learning & life is intriguing and there is a lot to learn from him. (This is another great interview with him.)
In this podcast episode Tyler talks about Jazz music, things he changed his mind about (amongst other things remote work), hiring talent, how competing at chess at a young age taught him humility, his views on the nurture vs nature debate, why he wouldn’t recommend independent thinkers to go into academia and his appreciation for Indian folk music.
🛌 Why Do We Dream? A New Theory on How It Protects Our Brains
One thing I learned from this article that was super surprising to me: Humans can learn to echolocate (the thing that bats do to orient themselves). In fact, our ancestors who have often been exposed to long periods of darkness during winters probably have been much better in this than any person today - so impressive how adaptive the human brain is! The other interesting thing is that the authors propose, that the reason we dream is that we need to provide stimulus to our visual system while we sleep in order to not loose that ability… I don’t know enough neuroscience to judge how plausible that is, but definitely an interesting thought I hadn’t heard about before.
In the ceaseless competition for brain territory, the visual system has a unique problem: due to the planet’s rotation, all animals are cast into darkness for an average of 12 out of every 24 hours. (Of course, this refers to the vast majority of evolutionary time, not to our present electrified world.) Our ancestors effectively were unwitting participants in the blindfold experiment, every night of their entire lives.
So how did the visual cortex of our ancestors’ brains defend its territory, in the absence of input from the eyes?
We suggest that the brain preserves the territory of the visual cortex by keeping it active at night. In our “defensive activation theory,” dream sleep exists to keep neurons in the visual cortex active, thereby combating a takeover by the neighboring senses. In this view, dreams are primarily visual precisely because this is the only sense that is disadvantaged by darkness. Thus, only the visual cortex is vulnerable in a way that warrants internally-generated activity to preserve its territory.
Food for Thought.
🤖 **Shameless self-promotion**: Together with my colleague Ole & Fenja from Denmark Statistics I published a report on the effects of large language models on the Danish labour market. Main takeaway: The exposure to LLMs will be very unequally distributed in highly developed countries such as Denmark. If you have any feedback/comments etc. happy to hear them
🗯️ Came across an interesting paper that claims prompting LLMs with true AND false examples of reasoning improves output. Have tested on a task I am currently working on but without much success - might be because the particular task is not just logic reasoning but also incorporates some kind of ‘subjective’ judgement. Curious to see if this technique will become state of the art or if this is just a context specific finding.
💻 So apparently a human writing one page on a computer produces more CO2 emissions through energy demand than ChatGPT writing one page… Would not have guessed. Found via @emollick.
Random Stuff.
🧐 An interesting/inspiring collection of general life advice/idea for what to work on. Also contains many great quotes from smart people, eg.:
Nate Soares:
In my experience, the way you end up doing good in the world has very little to do with how good your initial plan was. Most of your outcome will depend on luck, timing, and your ability to actually get out of your own way and start somewhere. The way to end up with a good plan is not to start with a good plan, it’s to start with some plan, and then slam that plan against reality until reality hands you a better plan.
🤓 learnprompting.org offers an extensive, open-source documentation on prompt engineering best practices - can recommend. Also recently listened to an interview with the founder - can also recommend.
🍔 Cost of a big mac at McDonalds across the U.S. Quite impressive that the most expensive one is more than double the price than the cheapest. Obviously had to look up ‘my local’ McDonalds - pretty much in the middle at a solid $5.99.
Personal Update.
As mentioned in the food for thought section (but maybe you skipped this one), together with my colleague Ole & Fenja from Denmark Statistics, I published a report on the effect of large language models on the Danish labour market.
Apart from working on finishing the above I have been exploring MIT & Boston. Some first impressions:
MIT has a so-called Banana lounge, i.e. a random classroom full with boxes of bananas. Anyone can just walk in at any time and take a free banana. Its financed by a donor. Probably the most popular MIT sponsor amongst all students.
Visited the Boston library - its beautiful! And anyone can just go and work/ study there
Managed to sneak into a couple of classes at MIT on topics such as Behavioral Economics + ML, Metascience & Augmenting Cognition - so far been really impressed by the quality of teaching, hope it stays that way.
Bought a second hand road bike. In retrospect I think it wasn’t really a good deal. In the spirit of learning from bad decisions (thanks Farnham Street): Note to self: if buying stuff second hand, write down a list of criteria beforehand to check when trying things out (I somehow always get nervous/irrational when having to make buying decisions on the spot/under pressure/when people are watching me).