Arjun Panickssery Profile picture
Researching scalable oversight @MATSprogram | prev @METR_Evals @ai_risks | spaced repetition | AI safety | https://t.co/mc28sVZYOC
Apr 7 5 tweets 2 min read
Mississippi has the best demographic-adjusted NAEP (4th & 8th grade) scores now

The "Mississippi Miracle" started in 2012 when the Republican governor/legislature introduced phonics-based instruction and began to hold back ~10% of 3rd graders per year who fail a reading test Image In contrast, Oregon, with the lowest demographic-adjusted scores, has a Board of Education that has indefinitely "paused" since 2020 the use of any standardized test as a graduation requirement

Most of this stuff isn't rocket science
Mar 31 7 tweets 2 min read
The narrative in which college admissions is inherently a rat race even without DEI/legacy-admissions/etc doesn't pass napkin math

1) 3.8m students graduate hs each year
2) Top-20 colleges admit ~40k total

So they could just admit a large majority if not all 1550+ SAT scorers? In 2023, of the 2.13m students who took the SAT, only 38k scored above a 1500 and only 8k scored above 1550

So they're not a dime a dozen; it's just that every single university chooses the majority of their class from students who score below 1560

Sep 11, 2022 5 tweets 1 min read
I often hear people say they increase their efficiency/impact by trading time for money, e.g. valuing their time at $35/hr and booking an Uber for $15 to save 30 minutes of walking as a result

But is this compatible with your marginal time value in most models of productivity? Extreme example: suppose that you can only work productively for 6 hours a day, then your time is worth $0 afterward and you should just walk

I guess under other simple models, non-Uber examples like administrative tasks exhaust some kind of motivational resource ineffectively