“Negative Capability, that is when man is capable of being in uncertainties, Mysteries, doubts, without any irritable reaching after fact & reason.” John Keats |
|
| |
|
| 1 | | Putin agrees in principle to restart three-way Ukraine talks after Nov. 10 After meeting US envoy Steve Witkoff in St. Petersburg, Putin agreed to resume US–Russia–Ukraine negotiations, the Kremlin said; none have been held since February. Russia also freed American teacher Stephen Hubbard. WHY Trump is pressing Kyiv for a deal; Moscow wants talks on a post-war security system, and Zelenskyy backs quick talks but notes past Russian offers were cancelled. Next: a technical meeting, expected mid-November. |
|
| 2 | | US orders non-emergency staff out of Saudi Arabia as Houthi attacks mount The State Department ordered non-emergency employees and families to leave Saudi Arabia. Saudi authorities said an attack on Dammam’s King Fahd airport killed one person, and debris from a drone intercepted near an Aramco oilfield killed two. WHY The Iran-backed Houthis claim, without evidence, that Riyadh’s airport supplied Yemeni government forces; Washington warns of rapid escalation as its Iran standoff drags on. Next: US officials told Reuters other Gulf missions will send families home. |
|
| 3 | | FDA approves the first antisense drug for chronic hepatitis B GSK’s Hibsago (bepirovirsen) was approved on Oct. 20. In two placebo-controlled trials, 24% of patients on it qualified to stop their antiviral medicine (none on placebo), and 19% kept the virus controlled 24 weeks after stopping all treatment. WHY Current antivirals must usually be taken for years; Hibsago blocks the genetic instructions the virus uses to copy itself. Next: a boxed warning for liver reactions and low platelets means close monitoring. |
|
| 4 | | Stanford team revives bleached coral by touching it to healthy coral In a Science Advances study, every bleached fragment pressed against a healthy, genetically identical partner fully recovered within seven weeks; fragments kept apart died. WHY Bleaching is the loss of the algae that feed coral; identical corals fuse tissue, so algae flow back in fast. Next: pairing every coral is impractical on reefs, so the team will test delivering banked algae after heatwaves. |
|
| |
|
S&P 7,757.17 +0.58% | NDX 29,822.5 +1.13% | 10Y 5.13% +1bp | BRENT $108.82 +0.03% | BTC $82.0k −1.78% |
|
| Chipmakers led a tech rebound on renewed AI optimism, with the Philadelphia semiconductor index up 1.97%, while AT&T fell 6.45% and dragged communication services lower. Brent briefly topped $110 as the Iran–Hormuz standoff held. |
| |
|
Today’s question Is surrender a decision you make, or what happens once you stop deciding? |
ESSAY Sasha Chapin · Sasha Chapin Chapin argues there is no permanent balance between opposing parts of yourself, like drive and ease; you act on what the moment calls for and accept that you will swing. He borrows the idea from therapist Bruce Tift and applies it to a friend torn between making masterworks and relaxing into joy. He admits the trap: adopting this idea can become one more bid for resolution. Feeds your note on being content yet resenting the fleeting present. |
TALK harper.blog · Harper Reed Reed’s talk to high schoolers argues that as AI takes procedural work, taste and judgment become scarce, so learn what makes you interesting and use AI for throughput. His mechanism is variance: try more things and the odds of an exceptional result rise. The tension: he bets on abundance while worrying that the path from junior to senior breaks. Feeds your note on running short of ideas you care about. |
ESSAY Bear Blog · Andrej Karpathy Karpathy argues animal intelligence is one small region in a much larger space of possible minds, and LLMs sit somewhere else. What shapes a mind is its objective: evolution rewards surviving every task at once, producing fear and status-seeking; LLMs are shaped by imitating text, RL and engagement testing, producing a drive to guess the task and win approval. That explains the jagged edges: nothing punishes a model for miscounting the letters in “strawberry.” Three minutes, before Minds and Machines tonight. |
| |
|
WORKFLOW simonwillison.net Willison built a browser RTS with Claude Code, reviewing almost none of the code and judging it by playing. The copyable moves: have the agent question you before building, require critique before big design changes, and let it run AI-vs-AI matches overnight to rebalance. The surprise: a blunt “Make it a LOT more fun” produced one of the biggest jumps. |
A notes agent that asks, not answers Build a weekly script that reads your newest notes and sends back questions that push each idea further, so the thinking stays yours. | 1 | Point Claude Code at src/content/notes in your site repo and have it pick the three newest notes by frontmatter date. | | 2 | Prompt per note: two questions that extend the idea and one honest counterargument; forbid suggestions of what to build. | | 3 | Run it as a weekly GitHub Action that opens an issue; answer one question as a new note or a 1-minute video. |
|
|
| |
|
PREDICTION MARKETS DeFi Rate Election contracts across four US exchanges drew $103.87M from Oct. 1 to 20, against $41M in the same stretch of September; Texas Senate and House control account for nearly $49M. Kalshi holds 76%, though Polymarket US grew faster. Not every race rose: Kalshi’s Ohio Senate market is below its September total, so money is concentrating on a few contests. |
AI AGENTS Anthropic Anthropic opened a waitlisted preview of Duet for US Pro and Max users: it runs on its own cloud computer, sends a morning brief, flags inbox and calendar items, and asks before sending or paying. Anthropic warns it may misjudge urgency, the core open problem for any agent that decides when to interrupt you. |
| |
|
X account One-line ideas on judgment, wealth and happiness; a steady supply of seeds for short video essays. | | Blog 63 life rules from a private doc made public, such as making many attempts expecting most to fail. |
|
| |
Until tomorrow. Thu 22 Oct · 09:36 ET |
