Given that today's AI has a jagged frontier - it is really good at some very hard tasks, but fails at some very basic ones - why is there the assumption that AI would be good at everything all at once?

Jagged 'superhuman' performance seems like a likelier near-term than true AGI

Infertility causes a deterioration of mental health and increases divorce, with no “protective” effects on earnings; yet willingness to pay for IVF without insurance is relatively low, from Sarah Bögl, Jasmin Moshfegh, Petra Persson, and Maria Polyakova nber.org/papers/w32445

״אֶסְקֶפִּיזְם (בעברית: בּוֹרְחָנוּת) הוא דרך להימנע מחיים לא נעימים או משעממים, במיוחד על-ידי חשיבה, קריאה או צפייה בפעילויות אחרות. אסקפיזם יכול להיות גם נטייה התנהגותית של הסחת הדעת לבידור או לפעילות דמיונית כדרך לברוח מהמציאות״.

לאסקפיזם יש השלכות קצת יותר מורכבות כשמבצע

ראיתי הערב את בת הזוג שלי חוטפת כינון ישיר ממכת'זית

אני מחפש תיעוד כדי להגיש תלונה במח'ש נגד מפקד תחנת תל-אביב

עמדנו על המדרכה, בדוכן של מחזקים, וחילקנו חולצות FCK BNGVR

שימוש במכתזית בכינון ישיר הוא סכנת חיים המשטרה השתמשה גם הערב באלימות אקראית ולא מוצדקת כדי להפחיד אותנו

#CausalTwitter , check out our new paper📜!
We provide theoretical justification for and tons of practical guidance on using negative controls for falsification tests for IV exogeneity.
I had the privilege of presenting this paper at the last EuroCIM.

WP🚨 with Dan Zeltzer Daniel Nevo Bar Weinstein & Walk.
If you used an IV, you likely ran a falsification or placebo test for exogeneity. We show these tests can be *too harsh*. They can flag issues even if the IV is exogenous. #EconTwitter #CausalInference
arxiv.org/abs/2312.15624
🧵>>

Summer readings all my books FREE DOWLOADABLE (the link in the first comment)

For one thing, the Hotelling model will be more literally applicable than anywhere else

Exciting news: Promoted to Associate Prof. (from tenured Senior Lecturer) at TAU Econ and will continue as a visitor at Stanford Econ next year! Grateful for everyone's support.

Falsification tests are often too harsh and people can find problems even when their IV is exogenous.
Come hear more in my #SOLE talk today at 10 am.

🗓️Fri 10am at SOLE: Oren Danieli presents our work on falsification tests for IV designs!

How to be as 'smart' as Auto-Regressive LLMs:
- memorize lots of problem statements together with recipes on how to solve them.
- to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem.
- apply the recipe blindly and declare

Join us soon for this webinar on digital innovations in health!

First paper from our lab on LLMs and emotional support (led by Joanna Z. Li): Skill but not effort drive GPT overperformance over humans in cognitive reframing of negative scenarios.

Short thread of the results ⬇️

osf.io/preprints/psya…

A student in my class shared this FT article on transformers & I thought it was very nicely done: ig.ft.com/generative-ai/

Horseracing AI models against written medical exams for humans is intriguing, but not super informative about clinical value.

New study on telemedicine finds broadly similar (and small) impacts on utilization, cost, and quality in the US as we found in Israel.
Not trivial given the many differences between the healthcare systems!
Health Affairs:
healthaffairs.org/doi/10.1377/hl…
JEEA:
doi.org/10.1093/jeea/j…