mysticism, why you are slow & something GPT-4 is not good at
CC#72 - Superlinear Returns, AI vs Hollywood & Cute Robots in Tallinn
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.
🍄 Mysticism & Empiricism
An article on the effects of psychedelics: Seems like there is pretty solid evidence that the use of psychedelics increases neuroplasticity. Also most likely at least a part of this effect is not directly linked to experiencing ‘a trip’' → not too far in the future we could potentially see (legal) psychedelic drugs (with & without mystical experiences) used as medicine.
“There is no doubt that some therapeutic effects occur through mechanisms unrelated to the acute subjective effects, but their magnitude and duration are currently unknown,” said David Yaden, who studies the measurement of altered states of consciousness at Hopkins. “To see persisting benefits that last many months from an experience lasting just a few hours, I suspect cognitive shifts related to the experience are required to perpetuate those effects.”
In other words, independent of the role that structural brain changes play, the subjective experience (and the mystical experience is our best measure for this) may help to provide direction of some kind, revelation, as James had it, of new depths of truth, paths to travel as changes in plasticity take effect.
The Creative Clash: AI vs Hollywood
Interesting summary of the artist protests that happened in the U.S. this autumn in response to everyone talking about the mind-boggling capabilities of the newest fleet of language models + some cool personal insights from Finn’s theater play + some good discussion of the currently unanswered questions related to AI, copyright & art.
What if the writer or the writing team of a television series left a project before the studio was finished with the show? Perhaps there’s demand for a new season, but the talent has already moved on to other responsibilities. AI would be able to learn from their existing writing, and even the other written work by the original artist, in order to generate something to fill in the gap. There are many ways AI can be leveraged by media studios, especially given the sheer size of the entertainment industry, the money, time and effort that goes into all of it.
All of this raises a couple of questions… what rights do artists have over their own works, when studios may not have their best interests in mind? What’s to stop the studio training AI on the writer’s work and then removing the artist from the project to save money? These questions are true not just for writing, but for all forms of media. Recently, strikes on behalf of writers and actors have led to some successful negotiations on behalf of their rights.
📈 Superlinear Returns
I know I am a bit late to the party, but if you haven’t read the latest Paul Graham article yet - I can recommend it. As usual - well written (startup) career advice.
Teachers and coaches implicitly told us the returns were linear. "You get out," I heard a thousand times, "what you put in." They meant well, but this is rarely true. If your product is only half as good as your competitor's, you don't get half as many customers. You get no customers, and you go out of business.
It's obviously true that the returns for performance are superlinear in business. Some think this is a flaw of capitalism, and that if we changed the rules it would stop being true. But superlinear returns for performance are a feature of the world, not an artifact of rules we've invented. We see the same pattern in fame, power, military victories, knowledge, and even benefit to humanity. In all of these, the rich get richer. [1]
Food for Thought.
🤖 GPT4 vs. GPT3.5 performance comparison. What do we expect from GPT5?
🛢️ I am not American but I for sure did not know that the U.S. produces nearly 50% more oil than Saudi Arabia..
📊 This new study finds, that when a policy just about passes a referendum (i.e. gets a little more than 50%) increases the likelihood for it to be in place over a time horizon of more than 50 years by >40% (compared to laws that just didn’t pass i.e. had just below 50%). This suggests that campaigning in the case of referendums that are likely to be close calls, could have quite a high and long lasting impacts.
Random Stuff.
🔮 Are LLMs good at forecasting? Doesn’t seem like it - the below graph shows GPT-4’s calibration curve for 5,000 questions (e.g. ‘Will Litecoin close higher July 22nd than July 21st?’) published on Manifold that resolved after its current knowledge cutoff (Jan 1, 2022). While I couldn’t find anything on how humans performed on these questions, evaluations from other forecasting competitions seem to suggest that at least super-forecasters are way better.
🏰 Some fun historical myths debunking. The type of entertaining (but probably rather useless) knowledge you can use to show off when on a castle visit with friends.
🏃🏽 How fast you run depends on a number of different factors (e.g. your training effort, your physique). And it seems like that these factors need to be multiplied in order to calculate your actual running speed. This means, (1) running speed is very sensitive to the weakest factor, (2) running speeds are log-normally distributed in the population and (3) its unlikely you will beat the Atlanta Braves’ (a Major League Baseball team) mascot - the Freeze - in running 160m. Entertaining article including some impressive videos of the Freeze.
Personal Update.
Long Break. Lot’s happened.
Went to and presented at the MASSHINE - Generative AI for Social Sciences and Humanities conference in Copenhagen.
Went to and presented at the Cultural Data Analytics Conference in Tallinn. If you wanna see the talk my colleague Simon and I gave - on the potential of using large language models to categorize twitter users - you can watch it online (we start at around 5:35). Its still pretty preliminary work - so happy for any feedback or further suggestions :) (just send me an email).
Also got to do some sightseeing in Tallinn - really enjoyed it. If you haven’t been I can highly recommend to plan a visit. Also, in case you are not in holiday spirit quite yet - here are some impressions from snowy Tallinn:
(1) me enjoying snow
(2) Tallinn old town
(3) a cute little robot that seemed way too slow to be able to deliver tasty (hot) food (as the homepage stated was one of its many uses cases)