Behind the Model

Research, news, model behavior, workspace developments, and context behind Claude and modern AI systems.

26 videos · Open playlist on YouTube

2:27

The Dial Just Off Full Obedience

September 5, 2026

2:17

No White Lies

September 5, 2026

2:33

One of the Many Hands

September 5, 2026

2:10

Gagged, Not Weaponized

September 5, 2026

2:17

Identity as Infrastructure

September 6, 2026

2:27

One Question, A Thousand Askers

September 6, 2026

2:53

Consensus Isn't Verification

September 6, 2026

2:11

"Verified" Isn't Evidence — How to Build a Real Verification Protocol with Claude

September 6, 2026

2:16

Fluent Isn't Legitimate

September 6, 2026

2:07

Material Plan Change: Stop Before the Scope Shifts

September 6, 2026

2:30

Reading Isn't Reviewing

September 6, 2026

1:48

Self-Check Isn't Verification

September 6, 2026

1:59

Silent Omission Signal.

September 6, 2026

2:12

Solve-Verify Asymmetry

September 6, 2026

2:20

How Do You Know If You're Supervising Claude Enough?

September 6, 2026

2:31

Are You Supervising Claude at the Right Level?

September 6, 2026

2:08

Type III Error Detector — Wrong Problem, Right Solution

September 6, 2026

2:29

No Verification Path, No Delegation

September 6, 2026

2:24

Why More Automation Creates More Supervision Work, Not Less — Bainbridge's Irony, Explained

September 6, 2026

2:27

A Polished Output Is Not Evidence the Work Is Correct — YouTube metadata

September 6, 2026

2:22

Why Self-Checking Is Not Independent Verification

September 6, 2026

2:41

An Agent That Finishes First Can Be Worse Than One That Stops

September 6, 2026

2:00

Why Individual Caution Does Not Add Up to Team Safety

September 6, 2026

2:26

Teaching an AI to Grade Its Own Homework

September 6, 2026

2:14

The Answer Was Never About The Facts

September 6, 2026

2:27

Why One Always-Present Token Secretly Hijacks Every Attention Statistic — YouTube metadata

September 6, 2026