"that" after "than" or "like"

What this neuron does

This neuron lights up on the word "that" when it refers back to something just said, especially after "more than" or "just like." It works more like a pointer to a previous idea than a plain word. Turning it up made the model's output drift into unrelated, repetitive 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

   NASA launch director Omar Baez said after the satellite reached orbit. "You can't call it any better than that."⏎The probe began its Wednesday space shot at California's Vandenberg Air Force Base

2. #2 peak

   well-being. Most people think being a dentist means putting your fingers in people’s mouths and looking for cavities, but there’s a lot more to it than that. Dentists work in lots of places outside the dental office, including hospitals,

3. #3 peak

   total would have died at the beginning of each recursion in any case. A slightly more functional function, if still not very good would be:⏎Please note that even that code is completely useless :P I just felt like correcting the other code (too

4. #4 peak

   cough a lot -- sometimes the cough sounds just like that of whooping cough -- and can have trouble getting enough oxygen if the infection is

5. #5 peak

   incident light meter lux→EV standard is all about? Is it just an average-this-probably-will-work value, or is there more to it than that? And if it is just an average, what knowledge is necessary<|endoftext|>from The

6. #6 peak

   come to snorkel over the turtles and watch them feeding before getting up close and personal while they swim up to the surface for air. What's even better than that? Looking at them in the eye as you're scuba diving! The turtles

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

   features not associated with the lung parenchyma; the tissues involved in gas exchange serve as the main sites of COVID-19–induced damage. That means removing airways and blood vessels, and then enhancing the pictures of what is

2. #2 peak

   designed to split the pool into two groups, one with five potential candidates, and the other nineteen. That’s why there are only five women and nineteen men, five people with blue eyes

3. #3 peak

   100,000 pregnant women from the general population and assume that the prevalence of Down syndrome is 1 in 500 pregnancies. That means that there would

4. #4 peak

   were recently published in PNAS⏎Our ideas of protein function have undergone a marked transformation with the discovery of an aspect of enzyme evolution called "moonlighting." That is, a protein can acquire a second (moonlighting) function without con

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.