the spot after a comma in a number list

What this neuron does

The neuron fires on the empty space right after a comma in a list of numbers or years, such as "1, 1, 2, 3" or "7, 1862", where another number is expected next. Turning it up made the model produce lists of items and options, a weak match.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

   1, 1, 2, 3, 5, 8, 1

2. #2 peak

   1, 1/4 turn clockwise and then back 2_______.⏎5.1

3. #3 peak

   02, 2012⏎Repairing bones with nanomaterials⏎(Nan

4. #4 peak

   0, 1, depending on the quadratic residue properties of modulo .⏎A famous result of Gauss

5. #5 peak

   9, 2012⏎Share Email Print⏎Kate Wong is an editor and writer

6. #6 peak

   7, 1862.⏎This navy blue infantry battle regalia is made out of

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

   Linda University a limited, non-exclusive right to make this publication available to the public. The author retains all other copyrights.⏎Rosenblatt, David M., "Music Listening as Therapy" (2015). Loma Linda University Electronic

2. #2 peak

   objectives, Multiple stakeholders, Decision-making, Flood management⏎Civil and Environmental Engineering⏎Akter, Taslima and Simonovic, Slobodan P., "A General Overview of Multi-objective Multiple-participant Decision Making for Flood Management"

3. #3 peak

   Constitution Act, 1982. Chapter 11 of the UFA establishes a blueprint

4. #4 peak

   , Proceedings of the 13th USENIX Security Symposium, 2004.⏎Garfinkel, S. and Shelat, A., "Remembrance of Data Passed: A Study of Disk Sanitization Practices," IEEE

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.