Inside the Black Box: Cracking AI & Deep Learning | Arshavir Blackwell, PhD | Substack

Inside the Black Box: Cracking AI & Deep Learning

Mechanistic Interpretability and Artificial Psycholinguistics in LLMs

By Arshavir Blackwell

No thanks

Inside the Black Box: Cracking AI & Deep Learning

The Bench Beneath the Benchmark

What the Active Site RCT tells us about fluency, capability, and the limits of evaluation

May 26•Arshavir Blackwell, PhD

4

LatestTopDiscussions

The System Prompt Is Not the Control Layer

A synthetic insurance pilot showed that policy prompts can make an AI sound more governed without making it safer.

May 16•Arshavir Blackwell, PhDJohn Holman, and Silvia Stepitova

8

The Internal–Output Decorrelation

Why What a Model Learns Inside Doesn’t Predict What It Does Outside

May 4•Arshavir Blackwell, PhD and John Holman

9

3

4

Breaking the Loop

Practical Epistemics for a Post-Fluency World

Apr 27•Arshavir Blackwell, PhD

9

Why Your Brain Believes the Model

The Heuristic Loop You Can't Break from Inside

Apr 20•Arshavir Blackwell, PhD

4

6

2

Fluency as Validity

How LLMs Make Thinking Feel Finished

Apr 13•Arshavir Blackwell, PhD

8

4

1

What Seneca Teaches Us That Marcus Couldn't

716 Stoic Features and Two Kinds of Inert

Apr 6•Arshavir Blackwell, PhD and John Holman

10

6

3

The Pattern Holds Across Another Author

Seneca’s Letters Support What We Found in Marcus Aurelius

Mar 30•Arshavir Blackwell, PhD and John Holman

4

1

See all

Inside the Black Box: Cracking AI & Deep Learning

Mechanistic Interpretability and Artificial Psycholinguistics in LLMs

Recommendations