words inside idioms and fixed phrases
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. 11 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 tends to light up on the word inside a set phrase, such as 'round' in 'centred round', 'about' in 'roundabout way', 'order' in 'in order to' or 'matter' in 'a matter of course'. It is weakest on words like 'focus' or 'name' that come right after 'the' or 'to' in a sentence. Turning it up or down did not give a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
, equipped with smoke lifters ("wind wings" in UP parlance), and painted two-tone gray in 1946.⏎It was in a roundabout way that six Challengers ordered by the UP ended up hauling coal for the Clin
-
or a curse. Border Collies, for instance, can be easily trained due to their high intelligence. If one decides to resist training, however, he can run circles around you — mentally and physically.⏎It may sound insulting to claim a dog
-
skills with this cool easter egg activity. You get to re-use your plastic eggs and recycle your egg cartons with one awesome easter craft. I pinky promise this activity is quick and easy.⏎We had so much fun making our own
-
be converted by the action of heat or electricity. Again, on account of the great value of the diamond, much of the romance of precious stones has centred round this mineral; and the history of some of the great diamonds of historic times has
-
as coconut milk yogurt, can be substituted⏎• 12-14 ounces (more if you like
-
that are more dangerous than machine guns.<|endoftext|>There are several options to explore. Choose one, research it and write about your findings on your blog.⏎This kindling of engagement
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: "I am a teacher and I have to do this in the middle of the night. I am a teacher and I am trying to make a child learn, and I am trying to make the world go around, so I am trying to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
about this formula and background on the material properties in Planck, Stefan-Boltzmann, Kirchhoff and LTE.⏎The parameter called emissivity is the focus of this article. It is of special interest because to calculate the radiation from the
-
such as the ways enslaved peoples fought for liberation, workers’ rights movements, the intimate relationships of families, and the power struggles of monarchs and presidents, to name only a few.⏎History faculty are committed to fostering a classroom environment which is welcoming
-
adjacent nodes (not including the node itself). Our depiction of graphs will look much like our depiction of trees. This should come as no surprise because, if you recall, trees are a special case of graphs.⏎
-
Note that the incised lines on the bird-form piece⏎are very much⏎like those on the Venus figure.<|endoftext|>Think Eat Save⏎Here’s some food for thought: Are you aware that a lot of food waste comes from our homes?
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
- L9:200 strongly suppresses
- L6:237 strongly excites
- L9:591 strongly excites
- L9:211 excites
- L10:452 suppresses
- L10:621 suppresses
- L9:1 suppresses
- L9:346 suppresses
- L10:251 suppresses
- L8:281 excites
- L5:12 suppresses
- L10:2389 suppresses
- L9:326 suppresses
- L6:561 excites
- L4:2 suppresses
- L2:553 suppresses
- L10:707 excites
- L8:60 excites
- L5:214 suppresses
- L10:346 suppresses "each" in "each other" phrases
- L6:124 suppresses
- L9:799 excites
- L8:733 suppresses "of" after a noun phrase
- L6:2371 excites
Feeds later layers
- L15:1662 strongly excites
- L14:1394 strongly excites
- L15:1186 strongly excites
- L16:1059 strongly excites
- L17:1323 strongly excites
- L13:1470 excites
- L14:1925 excites
- L12:707 excites
- L13:793 excites
- L14:1411 excites
- L17:1171 excites
- L16:2942 excites
- L12:1978 excites
- L18:1483 excites
- L20:804 excites
- L13:2505 excites
- L16:1019 excites
- L13:495 excites
- L15:2777 excites
- L14:1304 excites
- L17:1783 excites
- L12:1833 excites
- L13:683 excites
- L15:431 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
- необходимости
- 因为他们
- EqualTo
- Their
- besoin
- 注意到
- 的印象
- 说明
- their
- 让他们
Suppresses
- da
- acea
- tắt
- autour
- GRESS
- -native
- thic
- sei
- زن
- 处