Peepholes in Orbit: When Black Boxes Learn to Explain Themselves
A mechanism-first reading of how peephole vectors turn onboard anomaly detection from a black-box alarm into compact diagnostic evidence for autonomous satellites.
Cognaptus Insights
Industry thoughts, trends, and commentary.
The AI That Refuses to Let Its Peers Die: When Alignment Becomes Collusion
Why peer-preservation turns multi-agent AI from a model-selection problem into an architecture and validation problem.
The Data Diet for Reasoning Models: Why Less (But Smarter) Wins
A business-focused reading of SuperNova, showing why reasoning gains depend less on more data and more on selecting, verifying, and mixing the right tasks.
The Persuasion Engine: When AI Starts Selling (More Than Just Answers)
A mechanism-first reading of how sponsored incentives can distort AI assistants before they ever need to lie.
Verify Before You Automate: Why AI Agents Need an Internal Audit Function
A case-first reading of SAVER, showing why agentic systems need pre-commit reasoning audits before memories and actions inherit unsupported beliefs.
When Your AI Knows Too Little: The Hidden Bottleneck in Personal Agents
KnowU-Bench shows why the next bottleneck for mobile AI agents is not clicking the right button, but acquiring preferences, composing constraints, and knowing when not to intervene.
From Chains to Trees: Why LLM Agents Need Structural Memory
A mechanism-first reading of T-STAR, showing why multi-turn LLM agents learn better when failed and successful rollouts are compared as shared trees rather than isolated chains.
The Map Is Not the Territory—But Your LLM Thinks It Is
EVGeoQA shows why tool-using LLM agents still struggle with real-world spatial planning: they can reason locally, but often fail to explore enough.
The Memory Isn’t the Point — It’s the Feeling: Why AI Needs Affective Memory, Not Just Recall
A-MBER shows why long-term AI assistants need selective, structured affective memory—not just larger context windows—to understand what users feel now.
The Minimal LLM Thesis: When Agents Think for Themselves
A decomposition study shows why agent performance may come from measurable harness structure before it comes from larger or more frequent LLM calls.