surname pieces split across tokens

What this neuron does

The neuron fires strongly on the middle of a person's surname when the name is split into several pieces, such as 'Lo[ew]en', 'Ru[hr]' and 'Gu[ite]au'. Its strongest negative values appear on name pieces too, so it looks like a name-fragment detector with the sign depending on the case. Turning it up or down did not give a clear story in the text.

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

   eyes and a narrow short snout, similar to its relative T. rex, which lived 10-12 million years later. Dr. Mark Loewen, the study’s lead author, noted, “The width of the back

2. #2 peak

   Salt Lake City, Utah.⏎The study, funded in large part by the Bureau of Land Management and the National Science Foundation, was led by Dr. Mark Loewen, research associate at the Natural History Museum of Utah, and adjunct assistant professor

3. #3 peak

   words.⏎Professor Onur Güntürkün, one of the co-investigators from of Ruhr University’s Department of Biopsychology, says “that pigeons — separated

4. #4 peak

   can learn to distinguish real words from non-words by visually processing their letter combinations, surprising new research from the University of Otago in New Zealand and Ruhr University in Germany shows.⏎The researchers found

5. #5 peak

   the last months of WWII) against fnatical German resistance while US forces were encircling the Ruhr and preparing to cross the⏎Now a political war began. Churchill wanted Ike to

6. #6 peak

   easy for Charles J Guiteau to assassinate Garfield in 1881. Guiteau used a British Bull Dog revolver to shoot the President point blank in the back

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

   make<|endoftext|>Amaziah of Judah⏎Amaziah of Judah (אמציה) was a king of Judah, the son and successor of Joash, and the father of Azariah (Uzziah

2. #2 peak

   in any rush to get there.⏎Meet future woman: shorter, plumper, more fertile [New Scientist]<|endoftext|>MERCIER, JEAN ° (Joannes Mercerus ; d. 1570), French Hebraist.

3. #3 peak

   NC, Julio Cárcamo. However the current status of the project is unclear.⏎Main policy lessons⏎Tulio Monterroso, AMUPROLOAGO: mailto:firstname.lastname@example.org; email@example.com ;⏎Cohen,

4. #4 peak

   with coagulopathy have a threefold to fourfold greater mortality, and are up to

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.