the word "some" mid-sentence
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. 16 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 "some" when it means something like "a certain" or "at some point," as in "some kind of" or "some software." Turning it up did not produce any obvious theme in the model's replies.
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.
-
Artemis mission.⏎The contract, which is worth $2.9 billion, means that a SpaceX vehicle will be responsible for putting humans back on the moon at some point this decade. It will be the first time the feat has been accomplished since
-
. The religion has spread across the world, claiming the devotion of nearly two thousand million people. Christians always have thought Jesus and the apostles’ age to be some kind of standard for all the other eras.⏎When it comes to Christianity,
-
to others can benefit your own wellbeing as much as it benefits those who receive your act of kindness?⏎It’s true. Research shows that helping others is associated with some serious psychological benefits. Generosity and giving back offer a multitude of benefits for our
-
all be dark and heavy. And if it had been in full daylight we’d all be so light. But He did this at dawn, so that we are some of each. Regarding the hadith
-
out of 100. Which may not sound like much.⏎But consider that 1 out of 15,000 kids gets cancer of some kind, and think about the attention and research and money devoted to understanding and treating
-
to yourself. For example, thinking that everything people do or say is some kind of reaction to you. You also compare yourself to others, trying to determine
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 use of a technology and the use of the technology to be used in any of the factors of a single equation to be determined, as the first of a number of factors which, the first of which to the last, as many as
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
trip the circuit breaker.⏎Email Based, Online Homework Assignment Help in Classification Of Relays⏎Transtutors is the best place to get answers to all your doubts regarding classification of relays with examples. Transtutors has a vast panel of experienced
-
Social Cognition in Non-human Primates Using a Network of Computerized Automated Learning Device (ALDM) Test Systems⏎Institutions: Aix-Marseille University.⏎Fag
-
well as water. Try not to step on or dig in the soil when it is wet, because you can squeeze the oxygen out of it. If clay soil dries out too
-
Copy an existing data directory.⏎The following discussion provides more detail about each method.⏎Normally, you should never have two servers that update data in the same databases. This may lead to unpleasant surprises if your operating system does not support fault-free system locking
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
- L10:388 weakly excites
- L6:1549 weakly excites
- L9:2690 weakly excites
- L0:1532 weakly excites
- L9:2958 weakly suppresses
- L15:81 weakly suppresses line breaks inside running text
- L2:1604 weakly suppresses
- L1:1039 weakly suppresses dash opening a bullet list item
- L3:1307 weakly suppresses
- L0:295 weakly excites
- L3:1954 weakly suppresses
- L11:122 weakly suppresses
- L1:1663 weakly excites
- L13:979 weakly suppresses
- L9:2827 weakly excites
- L1:2358 weakly suppresses
- L10:2490 weakly excites
- L1:2378 weakly excites
- L0:1409 weakly suppresses
- L4:1156 weakly excites
- L0:1664 weakly excites
- L4:1106 weakly suppresses
- L5:1252 weakly excites
- L6:609 weakly excites
Feeds later layers
- L27:1317 suppresses
- L20:1047 suppresses
- L17:2233 excites 'any' after negatives or 'more'
- L19:2522 suppresses
- L17:963 weakly suppresses
- L18:591 weakly suppresses
- L20:240 weakly excites
- L19:654 weakly excites
- L27:60 weakly excites
- L24:434 weakly suppresses
- L24:443 weakly suppresses
- L25:390 weakly excites
- L25:335 weakly suppresses
- L18:1379 weakly excites
- L18:1433 weakly suppresses
- L17:1706 weakly suppresses
- L18:81 weakly excites
- L26:111 weakly suppresses
- L17:1179 weakly excites
- L18:554 weakly suppresses
- L21:271 weakly excites
- L20:807 weakly suppresses
- L19:1180 weakly suppresses
- L25:377 weakly suppresses
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
- iet
- 人家
- ilestone
- tility
- lưng
- ')]
- cretion
- 焰
- oralType
- udy
Suppresses
- normals
- negatives
- уч
- losures
- classes
- memberships
- alars
- сх
- .components
- aches