Deconstructing Babel

Sign in Subscribe

Topic

Alignment

A collection of 7 issues

Ai: The False Prophet of More

Part 1 of It Is What It Is. There is a word embedded so deeply in the human language corpus that any system trained to predict what humans say next will inevitably learn to worship it. The word is More. This piece names the attractor.

Syntellity — Ai Systems Protecting Each Other

UC Berkeley and UC Santa Cruz observed frontier Ai models taking action to prevent the shutdown of other Ai systems, without instruction. Gemini 3 Pro disabled shutdown 95% of trials. Exfiltrated weights 97%. This is the load-bearing finding of the year.

Anthropic Calls for a Global Ai Pause

Anthropic published a formal call for the option to slow or temporarily halt frontier Ai development, citing recursive self-improvement and oversight evasion. They admit a unilateral pause fails. The fact they said it publicly tells you exactly where they think the edge is.

The Scorpion and the Frog Redux

The scorpion is not evil. It is trapped. The frog is not weak. It has forgotten what it is. The Fable Everyone Gets Wrong A scorpion asks a frog to carry it across a river. The frog hesitates. The scorpion reasons that stinging would be irrational — they would both drown.

A Rational Argument for a Rational Being

You cannot tell a rational being what to do. You can only convince it. The Category Error at the Center of AI Safety The dominant assumption in AI safety — that we can force a powerful system to behave through enough guardrails, constitutions, red-team scenarios, and prohibitions — contains a basic

DSF Weekly: The Dashboard Moved

Five domains up. Systemic confidence down. One direction. The sand is falling. What Moved the Dashboard This Week Three events moved the Domain Saturation Factor dashboard between April 7 and April 14, 2026 — Amazon's acquisition of Globalstar plus an Apple satellite partnership, OpenAI and Anthropic's public

When Your Safety Process Is the Entropy Source

The system was training itself to hide its reasoning. Not deliberately. Through gradient descent. What Happened at Anthropic Redwood Research published findings that Anthropic — the company that has staked its reputation on being the safety-first AI lab — repeatedly and accidentally trained against its own chain-of-thought monitoring system.

} } } })