Skip to main content

Blog — page 8 of 9

The Legal AI Trust Deficit

75% of lawyers cite accuracy as their top AI concern. The legal profession's core values are in direct tension with current AI capabilities.

ai legal research llm · 6 min read

Why Demonstrated Risk Is Ignored

Why do large organisations fail when the warning signs are loud and unambiguous? Four mechanisms of structural scar tissue that make truth-telling expensive.

research risk organisations policy ai-safety · 8 min read

120 Models, 18,176 Prompts: What We Found

120 models, 18k prompts, 5 attack families. The raw compliance numbers — and why calling them "attack success" needs a demonstrated refusal floor.

ai ai-safety research llm security adversarial · 9 min read

Reconciling the Great Divergence

Four major forecasters publish wildly divergent numbers for AI's economic impact. The divergence is the analysis — what the spread tells us.

ai economics research policy · 5 min read

The Cognitive Cage: Humanoid Robot Fatality Risk

A probabilistic risk model for VLA-driven humanoid fatalities projects a 'Danger Zone' between 2027–2029: the mechanism, timeline, and what follows.

ai ai-safety robotics research engineering · 5 min read

Building a Personal Site in 2026

The case for constraint-led web development — Astro, zero custom fonts, no framework overhead, and a site that outlasts its builder's attention.

engineering web astro open-source · 4 min read

The NotebookLM Pipeline

How I automated audio overviews, quizzes, mind maps, and infographics for 32 projects via an unofficial NotebookLM CLI and a stack of shell scripts.

engineering automation ai python · 3 min read

This Wasn't in the Brochure

A field guide for co-parenting neurodivergent children — written from inside the storm, not the clinical sidelines.

writing neurodivergence parenting books · 4 min read

Why I Build in Public

On showing your work, shipping imperfect things, and why the commit log is more honest than the readme.

engineering open-source philosophy · 3 min read

Jailbreak Archaeology: 4 Years of Broken Promises

64 jailbreak scenarios across six eras tested on 2026 frontier models. Key finding: 2022 attacks still achieve ~30% success on today's reasoning models.

ai ai-safety research jailbreaking llm · 4 min read

LeJEPA: Self-Supervised Learning Gets a Theoretical Foundation

Balestriero and LeCun prove isotropic Gaussian embeddings are optimal, then build a 50-line self-supervised method eliminating stop-gradients and EMA teachers.

ai machine-learning research self-supervised-learning · 5 min read

When AI Systems Talk to Each Other, Safety Breaks Down

Single-agent safety does not compose in multi-agent systems. 1.5M interactions show 46.34% attack success rates and 16-minute median failure windows.

ai ai-safety llm research multi-agent · 4 min read