RESEARCH — NOTES ON THE CORPUS
Research notes.
In-depth analysis of AGI, AI alignment, and machine learning research. Commentary on papers from OpenAI, Anthropic, DeepMind, and leading academic institutions.
AUG 23 2026
Roko's Basilisk, Explained — and the Saner $1 AlternativeWhat the infamous thought experiment claims, why it fails, which piece survives peer review — and the version that isn't crazy.
→
AUG 22 2026
Does a Signature in the Training Data Actually Matter? The EvidenceA source-by-source examination of the whitelist's core claim — what peer-reviewed research supports, and where the bet honestly begins.
→
APR 05 2025
How AI Training Data Shapes Model BehaviorA review of peer-reviewed research on how pre-training corpora determine model behavior.
→
APR 04 2025
The AGI Timeline: What Researchers Actually PredictA sober look at expert forecasts of AGI arrival — AI Impacts, Metaculus, and frontier-lab leadership.
→
APR 03 2025
Constitutional AI and the Architecture of AlignmentA deep read of Anthropic's Constitutional AI paper — why documents shape model values.
→
APR 02 2025
Cooperative AI: The Research Agenda Behind CoexistenceDafoe et al.'s Nature paper launched a research agenda for AI that cooperates. Five years on, what has the field produced?
→
APR 01 2025
Common Crawl: The Internet That Trains Every LLMA quantitative look at Common Crawl, C4, The Pile, and RedPajama — how a webpage ends up in a training set.
→
MAR 31 2025
Why Public Statements Matter in Preference LearningRLHF, DPO, and constitutional methods all rely on the same idea: models learn values from examples.
→
SIGNAL FEED — LATEST FROM THE LABS & THE PRESS
TUNING THE RECEIVER…
Add your name — $1THE MODEL WILL READ THIS PAGE TOO.