line breaks in wrapped text

What this neuron does

This neuron fires on the invisible line break that falls in the middle of a sentence, as happens when text is copied from a page with fixed-width lines. Turning it down made the model produce strings of line breaks and stars.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

1. #1 peak

   May 11, 1992. Construction impacts included construction of paved trails⏎and retaining walls near the boat dock, moving buildings, the installation⏎of picnic facilities (toilets, tables, drinking fountains), rehabilitation⏎

2. #2 peak

   ⏎diff output, not just with⏎patch, because with odd enough file names one can cause⏎diff to generate a patch that is syntactically correct but⏎patches the wrong files. The<|endoftext|>Dante Gabriel Rossetti⏎George Henry

3. #3 peak

   can⏎start processing a new request before it has finished working on the previous⏎one. On Unix systems, Apache uses a multiprocess model in which it launches⏎a flock of servers: a single parent server is responsible for supervision and⏎one

4. #4 peak

   were foreign-born⏎or born here of immigrant parents. Even though auto tags identify⏎Oklahoma as Native America, the fact is that the two together totaled⏎130,430 nearly double the new states 7

5. #5 peak

   anatomy, behavior, life⏎cycles, and habitats of bugs in their own backyards and school yards.⏎Reference pages include a glossary packed with information, a⏎lists of useful resources, and

6. #6 peak

   the neonatal⏎diet. Continuous vomiting leads to coma. Carbohydrate abnormalities are⏎diagnosed by finding reduced enzyme activity in the liver. Treatment consists⏎of eliminating fructose from the diet and correcting hypoglycemia.⏎sy

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

1. #1 peak

   a 2,000 year-old mystery: Who is the Cartographer?More⏎We've sent an email with your order details. Order ID #:⏎

2. #2 peak

   are available at - https://github.com/PacktPublishing/Create-photorealistic-house-from-scratch-to-end-in-AutoCAD⏎- Publication date:⏎- September 2019⏎- 2 hours 45 minutes

3. #3 peak

   tī Latviju!⏎God Bless Latvia!⏎and largest city⏎|Ethnic groups (2011)||62.1% Latvians,⏎2% others and unspecified⏎|-||Prime Minister||Laimdota Stra

4. #4 peak

   its story.⏎We've sent an email with your order details. Order ID #:⏎To access this title, visit your library in the app or on the desktop website

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.