"than" after "more to it" comparisons
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 6 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word "than" in phrases like "there is more to it than" or "goes a lot further than", where something is being said to be more than a simple version. Turning it up or down did not give a clear change in the written 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 activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
and 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
-
under domestic trade rules are permissible and viable options according to international norms, but showing favoritism for one industry at the expense of others can do far more harm than good. It is time for the U.S. government to stop implementing special interest
-
the 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
-
While some may question how much solving six problems of their choosing could possibly be of much benefit to students, there's actually a whole lot more going into this assignment than just solving six math problems. Students become familiar with the layout of a newspaper,
-
the environment in a number of ways1?⏎1) Cut Your Greenhouse Gas Emissions⏎The impact of animal agriculture on greenhouse gas emissions goes a lot further than just cows producing methane gas. Meat production requires vast amounts of energy. Not only
-
⏎There is more to mantis shrimp than meets the eye. They pack a punch, decimating their prey using deadly strikes
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
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.
Here is something I have been thinking about lately: if I can have a lot of people in the group, and if I can have a lot of people in the family of the people of the country. (The first time and the second time) of the said problem in the family.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
nickname of “imperial city”, envied by kings like Charles V. The expression “Vale un potosi!” (it costs a potosi
-
ser to the Sun doesn’t mean warmer⏎Mercury is the closest planet to the Sun. So it would make sense that this is the hottest planet,
-
Classes define the behavior for objects of its type. Instances of the class Class represent classes and interfaces in a running Java application. If we do not explicitly write a constructor for a class, the Java compiler builds a default constructor for that class. so
-
well as technical. Most employer surveys do, however, note that personal skills have at least equal rating with technical skills or qualifications for example;⏎In the Confederation of British Industries Education and Skills Survey: 2015, employers state:
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L5:1580 strongly excites
- L2:1774 excites
- L4:221 excites
- L2:7 suppresses
- L1:55 excites
- L4:646 weakly excites
- L4:1654 weakly suppresses
- L1:2723 weakly suppresses mid-word syllables in capitalised names
- L4:1591 weakly excites
- L2:3006 weakly excites
- L1:721 weakly excites
- L0:2622 weakly suppresses
- L5:2165 weakly suppresses
- L4:347 weakly excites
- L0:3014 weakly suppresses
- L2:132 weakly excites
- L1:2030 weakly suppresses
- L0:2100 weakly excites
- L0:1811 weakly excites
- L4:678 weakly excites 'have' after 'you' or 'to'
- L3:2103 weakly suppresses
- L0:2725 weakly suppresses
- L1:2444 weakly suppresses
- L0:2863 weakly suppresses
Feeds later layers
- L10:559 suppresses
- L7:1324 suppresses
- L7:2203 excites
- L10:1384 excites
- L13:1213 excites
- L11:981 suppresses
- L19:1615 suppresses
- L7:1280 suppresses
- L7:1944 excites
- L23:2163 suppresses
- L26:850 weakly suppresses
- L9:1890 weakly suppresses
- L22:2334 weakly suppresses
- L14:643 weakly suppresses
- L8:1351 weakly excites
- L18:666 weakly suppresses
- L8:919 weakly excites
- L9:654 weakly suppresses
- L7:2716 weakly excites
- L20:931 weakly suppresses the digit 6 inside years
- L21:2276 weakly excites
- L26:2072 weakly excites
- L23:2135 weakly excites
- L10:1008 weakly excites
Activation histogramA chart of how often a neuron takes each activation value.
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 lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
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 layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- 烷
- ipse
- na
- ונה
- egal
- ⏎
- ⏎ ⏎
- 科室
- pane
- WiFi
Suppresses
- [value
- ehr
- issy
- 직
- -repeat
- Enough
- kehr
- cropped
- veau
- bem