Saturday morning links: scanning articles on the future of AI and society

Classes are nearly done. I give my Baylor exam next week, and gave my A&M exam this week. I am behind on sleep again for some reason too but I wanted to get this started and then maybe finish it with my morning coffee.

A new book argues dating apps have caused permanent dating, as opposed to dating as an onroad to long term relationships. Sounds horrible.

You can buy a house in parts of Italy for a dollar as part of a program by the government to revitalize inventory. But you’ll need to put some elbow grease into fixing them up so be sure to multiply your $1 investment by tens of thousands.

Poor measurement is one of the main problems facing AI at the moment. As in which large language model writes the best code. But honestly, at the same time, given their advances it would be like for me asking a series of judges to rank which students in my classes are the 45th to 55th percent tile. Something about this work and it’s quality makes it hard to rank, just like it’s hard to rank the work of workers when they’re neither some of the best or some of the worst at their jobs.

Climate change will have many costs, many of which aren’t yet counted. More hot days in some industries increase the risk of an accident for instance. But if there’s an increased risk of accidents, what will that do to the capital-labor ratio in those jobs and what will happen to hazard pay?

David Brooks makes a distinction between “resume virtues” and “eulogy virtues”, the former being things we take to the market and the latter others bring to our funeral. This new article on what’s happening with men referenced it and caught my attention.

A principal is facing prison time for intervening between two students in a way where while there was physical contact, it doesn’t appear there was assault by the principal. Growing up in a town where principals and teachers used corporal punishment, it’s jarring to read and contrast with my own education. Though I fortunately was never spanked.

Artificial intelligence wearables are probably the imminent future. Many are trying to get us off our phones. The Rabbit R1 apparently is better than Humanes AI pin. But neither will come close to Vision Pro which has ChatGPT as a downloadable app, and is wearable, so I’m going to call it a wearable AI device.

HBO’s sequel to “The Jinx” is out but I haven’t seen it. But this is the day I post links so I can close them from my tabs so I’m telling you about its existence.

Meta AI is available inside WhatsApp and I think Facebook and the claim that it’s a ChatGPT app killer is ludicrous in my opinion. Why? Precisely because of that — switching large language models is increasingly going to “feel like murder” as we increasingly and inevitably anthropomorphize them. See my earlier substack about memory and how all of these are going to make switching costs astronomically high — maybe as high as the cost of putting down a pet, I bet.

Zach Snyder’s sequel to Rebel Moon is on Netflix and as i didn’t see the first one, Im not eager to see this one. But apparently there is a huge camp of Snyder fans so they probably have seen it. And if you’re one, you don’t need this post, and if you aren’t one, you don’t care.

Some are lamenting the good old days of TikTok when it apparently had fewer commercials. I got on it last night for the first time in over a year (my friend in the state dept flat out telling me it’s spyware caused me to delete it), and I can’t say I miss it.

The art of packing the carry-on. I also don’t check luggage though this summer I’m in Europe for 2 months straight and I guess may have to change that rule. Haven’t decided yet.

The Guardian published a letter with a headline claiming mental health professionals are skeptical that MDMA therapy will be an effective treatment for some anxiety disorders. Keep in mind though there is already substantial enough evidence for it as an effective treatment of PTSD from a series of Phase 2 and 3 trials that were promising enough that the FDA gave it breakthrough status and is currently reviewing a new drug application from the non profit formerly known as MAPS. The evidence from medical trials, conditional on their selection criteria (ruling out those with bipolar and psychotic disorders for instance) consistently finds that MDMA therapy has shortrun effects of so far maybe up to six months (plus or minus) of alleviating chronic PTSD symptoms. (Nature pub). Like it or not, many classic mind altering recreational drugs are coming off Schedule I due to medical discoveries about their efficacy for treating a variety of maladies for which current technologies have been ineffective, like PTSD.

Rob Donnelly and coauthors including Susan Athey have a publication that’s now a few years old about counterfactual inference across product choices. Whats it about? Here’s ChatGPT giving me the explanation of the abstract written for an intelligent layperson with some causal inference background. Rob is the first PhD economist, btw, I know of who OpenAI has hired. But maybe there’s others. Rob was I think an Athey student at Stanfords doctoral program in economics.

This paper introduces a new method for understanding consumer decisions across multiple product categories, considering that a consumer's total satisfaction is the sum of their satisfaction from each category and that their preferences for product features and price sensitivity vary across products and are interconnected. The authors adapt advanced probabilistic matrix factorization techniques from machine learning to account for changes in product attributes over time and product availability, like items being out of stock.

To evaluate the model, they used data from periods of price changes or when products were unavailable, finding that their approach outperforms traditional methods that treat each product category separately. The model excels at predicting individual differences in consumer preferences and identifying which consumers are more sensitive to price changes, leading to more effective personalized marketing strategies. This shows that a more sophisticated analysis of consumer behavior can significantly enhance the effectiveness of targeted discounts and promotions.

David Blei is a computer scientist and coauthor on a transformer paper with Athey, speaking of, where they trained a LLM on 20+ million online resumes and then fine tuned with representative datasets like the PSID and NLSY. It outperformed other benchmarks at predicting people’s career trajectories. Transformer models are powerful predictive models despite the fact they are designed for mimicking intelligent speech. The attention mechanism is probably a major breakthrough in more ways than we realize. I wrote about the paper recently.

A godfather of the internet wrote a book with several arguments, long ago, for why we should delete our social media accounts. One of the gifts a former friend gave me was helping me realize through my personalized lessons to delete my social media accounts and I have to admit, I am grateful to her for it. The best lessons are often the hard ones.