'much' or 'many' after 'not' or 'little'
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. 7 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 responds to words like 'much', 'many' and 'little' when they follow a 'not' or a 'have not changed' style phrase, meaning a small amount. Turning it up did not produce text that clearly shows this idea.
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.
-
52, if you wanted to create a list of users, you would create a new variable and:⏎Now with the List data type, things haven’t changed much. To create a List of Users, you create a new variable and:⏎The
-
add another 21.5 cents per gallon. In addition, many local governments levy additional sales and other taxes on gasoline. Gasoline taxes have not risen much as the price of gasoline exploded upward.⏎The price of gasoline is much lower than
-
one hope beyond the grave; this was God’s promise to Abraham. God gave Moses the Law while Israel wandered in the desert. It was rigid. There was little flexibility in its demands. It declared “an eye for an eye and a tooth
-
1941, the first graduation of doctors of our institute was held, young specialists immediately after graduation went to the front to save lives. Unfortunately, not many of them managed to return.”⏎The participants of the studio “Door to
-
, not supposed to have a head. Ordinary people are just bones. The spear is the novelty, nevertheless still the gestures are ritual symbolism.⏎We do not know much about the rule of the Romans in the provinces. However we know that they in
-
Digital Download 7 Day Vegan Meal Plan | Digital download⏎Awesome Benefits Of The Paleo Diet⏎The idea behind the paleo diet is that we haven’t changed much since the time of the earliest humans. It is said that “From a genetic
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: the number 13. It is a true statement. The question is whether such a thing as a "thing else" exists. So, for example, if the same location, time, and date, and only if the same as
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of the container. This usually occurs in about 1-2 hours depending on the size of your jar. A full melt
-
ans' donations.⏎The eagles typically incubate their eggs for about 35 days, and this year on the EagleCam website the DNR
-
Bay roughly 30 percent of the setnet permit holders are women. NOAA Fisheries scientists from
-
the Study of Religion" by Dr. Clayton Crockett. In it<|endoftext|>Approximately 1.6 million people die worldwide every year as a result of pneumococcal infection, which
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
- L5:289 strongly suppresses
- L5:1087 suppresses
- L4:656 excites
- L4:3049 suppresses
- L5:3048 excites
- L6:1132 excites
- L4:1678 suppresses
- L6:222 suppresses
- L5:2545 excites the zero in round numbers like 100
- L6:2805 weakly excites
- L4:955 weakly excites
- L5:218 weakly suppresses
- L6:2646 weakly suppresses
- L4:1543 weakly suppresses
- L1:956 weakly excites
- L6:1116 weakly excites
- L6:572 weakly suppresses
- L1:1289 weakly suppresses
- L5:2787 weakly suppresses 'and' in 'more and more' pairs
- L6:2033 weakly excites
- L2:2714 weakly excites
- L2:2193 weakly suppresses
- L0:2517 weakly suppresses
- L5:2100 weakly excites
Feeds later layers
- L16:1273 strongly excites
- L12:1155 strongly excites
- L9:230 excites
- L8:1643 excites
- L14:1875 excites
- L10:82 suppresses
- L12:494 suppresses
- L12:1296 suppresses
- L9:781 excites
- L9:620 excites
- L11:1674 excites
- L13:2393 excites
- L18:623 suppresses
- L11:640 excites
- L18:940 excites
- L10:644 excites
- L11:922 excites
- L12:2068 excites
- L15:2651 suppresses
- L11:1106 excites
- L21:542 excites
- L19:1393 excites
- L8:1049 excites
- L16:1720 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
- glm
- ilit
- /security
- itez
- spar
- iards
- ris
- diff
- security
- bé
Suppresses
- -equiv
- valid
- full
- _VERTICAL
- stalls
- state
- 奋斗目标
- ận
- PostBack
- fate