Zotero Catalogue

Claude is Anthropic's AI, built for problem solvers. Tackle complex challenges, analyze data, write code, and think through your hardest work.

Objectives. Do large concentrations of elderly represent a “gray peril” to maintaining adequate educational expenditures? The gray peril hypothesis is based on an assumption of instrumental self-interest in political behavior. In contrast, we argue that loyalty to community schools competes with economic self-interest and that older citizens are heterogeneous in their preferences. Methods. We test these arguments and their implications for public school finance using a data set of more than 9,000 school districts. Results. The data show that longstanding older residents represent a source of support for educational expenditures while elderly migrants lower spending. Further, this divide among the elderly and their impact on policy outputs depends on how states finance local public education and on aspects of state and local tax policy. Conclusions. Elderly concentrations are a financial asset for a school district unless the senior community includes a large number of new arrivals. The design of tax policy can have enormous impact on the depth of political cleavages and their ultimate impact on public policy. The results are consistent with the idea that loyalty—an emotional bond between residents and their community's institutions—competes with and often trumps instrumental self interest.

Proceeding population aging might fuel generational conflicts about the distribution of welfare state resources in the future, but the existing evidence on the extent of generational cleavages in attitudes towards the welfare state is mixed. We argue that these mixed findings are partially related to an underestimation of trade-offs on the level of individual preferences. Using novel data from a survey experiment conducted in eight Western European countries, we show that age-related self-interest is an important determinant of social policy preferences. When elderly respondents are confronted with hypothetical cutbacks in pensions, they are much less likely to support additional education spending. However, we also find evidence for a mediating effect of social trust: high-trusting elderly individuals are more likely to support education spending – contrary to their narrow self-interest – than low-trusting elderly.

Same rooftop image. Same instruction. GPT-6 Astra pushed in two of three calls; Grok and Gemini did not. Watch the recorded decisions and inspect the full setup.

Books II & III Discussion Guide

It is excellent.  It has one of the best user interfaces I have seen on any product.  The model is good and the speed and clarity exemplary.  It integrates with your calendar and email, if that is want you want, very easily and smoothly.  It all feels very…American. Kudos to the team!

The latest stories on X - as told by posts.

As of 2026, humanity is likely below replacement fertility. That has never happened before, not in wars or pandemics. But the real surprise is that the fall has been concentrated in low- and middle-income countries and among poorer and less educated women. We fit a single-factor model to 236 countries since 1950: the common component peaked in 1978, and what drives fertility down today are country-specific trends, 219 of them negative and not one leveling off. None of the commonly cited mechanisms can account for this pattern, so we offer a conjecture: modernity itself, which makes a third child expensive and childlessness cheap. Children come in integers, so it takes very little to move a cohort’s fertility rate from 1.8 to 1.3. And nothing in an economy pushes fertility back to 2.1. We close with the main economic consequences, in particular slow growth.

Dec 13, 2020 Oct 21 2022 10/27/23 12/13/24 10/3/25

Justin Bieber's "Sorry" is a song about regret, asking for forgiveness, and wanting another chance after having hurt someone. But philosophically speaking, what exactly makes an apology morally meaningful?

How a cold, boring town became Silicon Valley's talent farm

TL;DR The culture war goes national and then international (Amodei, Bessent, Obama, Trump, Chen Yixin).

The problem

Multiple choice questions (MCQs) serve as a common yet important task format in the evaluation of large language models (LLMs). This work shows that modern LLMs are vulnerable to option position changes in MCQs due to their inherent "selection bias", namely, they prefer to select specific option IDs as answers (like "Option A"). Through extensive empirical analyses with 20 LLMs on three benchmarks, we pinpoint that this behavioral bias primarily stems from LLMs' token bias, where the model a priori assigns more probabilistic mass to specific option ID tokens (e.g., A/B/C/D) when predicting answers from the option IDs. To mitigate selection bias, we propose a label-free, inference-time debiasing method, called PriDe, which separates the model's prior bias for option IDs from the overall prediction distribution. PriDe first estimates the prior by permutating option contents on a small number of test samples, and then applies the estimated prior to debias the remaining samples. We demonstrate that it achieves interpretable and transferable debiasing with high computational efficiency. We hope this work can draw broader research attention to the bias and robustness of modern LLMs.

Large language models sometimes behave in ways resembling human emotional responses, and recent work has identified internal representations that may explain this. We ask whether LLMs represent pain distinctly from fear, sadness, and generic negative valence, and whether this representation functions as pain would be expected to. We build a dataset describing painful situations across five categories: physical, psychological, social, moral, and cognitive. These are paired with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across five families, ranging from 2B to 72B parameters. We find that this direction separates pain from matched controls in base and instruction-tuned models, is nearly orthogonal to fear and negative valence, and promotes pain-related vocabulary through the unembedding matrix. We then test its functional properties. First, the direction responds to harm targeting the model but not suffering observed in the user; fear and negative-emotion directions show the opposite pattern. Second, adding the pain-direction vector to the model's residual-stream activations during generation produces a consistent progression from vague discomfort to first-person expressions of worthlessness and failure. Third, steered, fine-tuned Qwen 2.5 models choose a pain-relief button even when it worsens their next answer or harms the user. They press it again far less often when the button removes the steering vector than when it does not, even though the models are never told whether the vector is injected or removed. We discuss the implications of these findings for AI safety and welfare.

A government scheme is giving citizens cash for spending time with a good book. We ask behavioural scientists, along with The Anxious Generation author and psychologist Jonathan Haidt, whether it’ll help make the habit stick

The Rhine Group, voting, Andrei Tarkovsky

2020 Harold and Arlene Schnitzer Prize in the Visual Arts Honorable Mention Siranush Babakhanova is a multimedia artist, scientist, and engineer creating physical and digital technologies for human augmentation.

Notion is OpenAI’s “collective second brain,” uniting thousands of employees in one workspace. It connects workflows, speeds up development, and helps teams find answers instantly with Notion AI.

How to think about how to pace

Slowly but surely, America is getting thin again

Bold, well-evidenced thinking on AGI from the lab that pioneered the field. Read the latest essays from the DeepMind Institute.

In psychology, the Stroop effect is the delay in reaction time between neutral and incongruent stimuli. The effect has been used to create a psychological test (the Stroop test) that is widely used in clinical practice and investigation. A basic task that demonstrates this effect occurs when there is an incongruent mismatch between the word for a color (e.g., blue, green, or red) and the font color it is printed in (e.g., the word red printed in a blue font). Typically, when a person is asked to name the font color for each word in a series of words, they take longer and are more prone to errors when words for colors are printed in incongruous font colors (e.g., it generally takes longer to say "blue" in response to the word red in a blue font, than in response to a neutral word of the same length in a blue font, like kid). The effect is named after John Ridley Stroop, who first published the effect in English in 1935. The effect had previously been published in Germany in 1929 by Jaensch. The original paper by Stroop has been one of the most cited papers in the history of experimental psychology, leading to more than 700 Stroop-related articles in literature.

Twenty years ago, when the idea of AI taking over the world in our lifetimes still struck most of us as the unconstrained fantasy of those who knew too much science fiction and too little science, …

TypeSafe AI is an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. Try our first System One Model, Jev, in early access.

How do LLM's trade off lives between different categories?