Day 1 · 1 April 2026
How did the United States hand over its future to corporations?Inkhaven: 30 Days of Posts
A 30-day writing challenge at Lighthaven, exploring AI safety, politics, and positive futures.
Day 2 · 2 April 2026
The Physics of Great StorytellingDay 3 · 3 April 2026
The Garden (Part 1)Day 4 · 4 April 2026
The Garden (Complete)Day 5 · 5 April 2026
Spectator SportDay 6 · 6 April 2026
How Does an Agent with Multiple Goals Choose a Target?Day 7 · 7 April 2026
Spend Money to Buy OpportunityDay 8 · 7 April 2026
Revisiting GSM-Symbolic: Do 2026 Frontier Models Still Fail at Confounded Grade School Math?Day 9 · 9 April 2026
This Is Not Financial AdviceDay 10 · 10 April 2026
How South Africa's Electricity Catastrophe Was (Mostly) FixedDay 11 · 11 April 2026
How to Make a Problem DisappearDay 12 · 12 April 2026
Is the Adult in the Room with Us Right Now?Day 13 · 13 April 2026
Characterising the Views on Safety from Frontier AI Labs: Anthropic and DeepMindDay 14 · 14 April 2026
Characterising the Views on Safety from Frontier AI Labs: OpenAIDay 15 · 15 April 2026
A Case for Persona Robustness as a Research AreaDay 16 · 16 April 2026
You Can Just Build Things with Claude CodeDay 17 · 17 April 2026
Do Models Know They're Lying When Claiming Fake Identities?Day 18 · 18 April 2026
Which Information Sources I Listen to the MostDay 19 · 19 April 2026
People That Inspire MeDay 20 · 20 April 2026
Contributing to Technical Research in the AI Safety End GameDay 21 · 21 April 2026
Giving AI a ChildhoodDay 22 · 22 April 2026
Whatever I Was Doing in 2020, I Should Do AgainDay 23 · 23 April 2026
Whatever I Was Doing in 2020, I Should Do Again (Continued)Day 24 · 24 April 2026
Do Emergently Misaligned Models Believe What They Say?Day 25 · 25 April 2026
The Great Smoothing OutDay 26 · 26 April 2026
Why Are Anglophone Countries Unhappier Than Their European Counterparts?Day 27 · 27 April 2026
From Output to ConnectionDay 28 · 28 April 2026
Attachment Theory Is Extremely Cool and UsefulDay 29 · 29 April 2026
Q&ADay 30 · 1 May 2026
Linkpost: Sanity-Checking 'Incompressible Knowledge Probes'