INDEX

Explanations

values, costs, and rewards

The neuron fires on technical reinforcement‐learning contexts—spotting terms like “reinforcement,” “state,” “reward,” “observed,” etc., common in RL papers and code.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

vated

-0.82

 binding

-0.81

Policy

-0.78

 Duplex

-0.77

PPM

-0.77

 keres

-0.76

้าม

-0.75

 bidang

-0.73

 policy

-0.73

 Tamaño

-0.73

POSITIVE LOGITS

 reward

1.38

reward

1.09

 rewards

1.05

rewards

0.99

 Reward

0.96

 punishment

0.90

 encouraging

0.89

Reward

0.89

 rewarding

0.87

 score

0.85

Activations Density 0.003%