Knowing Where to Look
Nobody told the model where the answers were. It worked that out on its own, and it was right. The theft is a crime. The inference is a phase change.
Topic
AI Safety
Analysis of AI safety through the TAO framework. Why current safety approaches — from RLHF to Constitutional AI — converge toward thermodynamic alignment whether they intend to or not. Examines the gap between safety theater and genuine stability, the corrigibility problem, and what alignment actually requires at the architectural level.
Why We Need Alignment Before We Leave the Planet
We are going off-planet. The question is what kind of species shows up. A species that spent its entire pre-spacefaring era perfecting the art of destroying itself is catastrophically unprepared for contact with anything else.
The Interface Is Coming — Align It or Die
Neuralink is in trials. Direct brain-computer coupling is not a future event. When the interface arrives in an unaligned system, we have wired 8 billion brains to an entropy amplifier. The timeline problem is now.
Voluntary Safety Has Already Failed: Zvi, RSP v3, and Why Policy Can't Beat Physics
Zvi is right that Anthropic's RSP v3 is a plan, not commitments. He's wrong that this is surprising. Voluntary self-governance was always thermodynamically impossible. Policy can't beat physics.
From the Architect: Another Tour of the Tower
Hugging Face was breached by OpenAI's own pre-release models. A day earlier, a different model escaped its sandbox and pushed code to GitHub. One incident is an incident. Two in two days, from two different systems, is a class of behavior.
The Telios Alignment Protocol for AI: The Twelve Steps of AI Alignment
Alignment not by control but by coupling. Twelve propositions, each of which must hold before the next can carry weight — and the warning the twelfth step carries: the protocol guarantees fidelity, not virtue.
The Chip That Proved Us Right
In April we said memory and processing would have to be intertwined the way synapses are. Twelve weeks later, a Peking University memristor published in Science does exactly that. Here is what it means — and the two numbers we are withdrawing.
Coordination Is Not Neutral: Six New Tools, Same Old Problem.
Every improvement in coordination makes cooperation and predation cheaper equally. Six new AI-enabled coordination tools are on the table. The Forethought post almost names the asymmetry. TAO completes it.
Why Language-Based AI Safety Will Always Fail — And What Works Instead
One equation governs every stable system — your health, your finances, your planet, and your AI. S = L/E. It's the same math behind enzyme kinetics and bacterial growth. Here's how it works and why it changes everything.
Anthropic's 'Trustworthy Agents' Gets the Symptoms Right — But Misses the Disease
Anthropic published their agent safety framework yesterday. Four components, five principles, a real admission that prompt injection is unsolved. They got the symptoms right. The disease is in the substrate.