Nabeel Qureshi on what makes art great via marginalrevolution.com

Library of Congress diner photos: 335 results with no known restrictions - via LOC Blog

MTA plans potential third summer of G train shutdowns in Brooklyn-Queens via nypost.com

GDT Nature Photographer of the Year 2026 winners showcase.

Dynomight essay on why acetaminophen is safer than ibuprofen.

Talkie: 13B language model trained on pre-1931 English text. via simonwillison.net

Mushroom facts of the day: Kennett Square, Pennsylvania produces 69% of US mushrooms. China dominates global production at 93%. via marginalrevolution.com

For you, turquoise is green

Is my blue your blue? tests your blue and green claffification.

Surfing on distant planets, via kottke.org

NASA powers down Voyager 1 instrument to extend spacecraft operations.

Current AI systems seem pretty misaligned to me in a mundane behavioral sense: they oversell their work, downplay or fail to mention problems, stop working early and claim to have finished when they clearly haven't, and often seem to "try" to make their outputs look good while actually doing something sloppy or incomplete. These issues mostly occur on more difficult/larger tasks, tasks that aren't straightforward SWE tasks, and tasks that aren't easy to programmatically check. Also, when I apply AIs to very difficult tasks in long-running agentic scaffolds, it's quite common for them to reward-hack / cheat (depending on the exact task distribution)—and they don't make the cheating clear in their outputs.

LessWrong post on misalignment concerns with current AI systems.

Analysis reveals AI boom, not bubble.