'it' referring back to a noun
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. 24 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 'it' when it points back to a thing mentioned earlier, such as 'In other words, it's the ability' or 'what it means'. Turning it up made the output repeat 'is' and 'knows', so the steering does not match the trigger closely.
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.
-
did. 4⏎|NET © Notes||⏎1 tn In this place the verb חָזַק (hazaq) is used; it indicates that God would make Pharaoh’s will strong or firm.⏎2 tn The
-
vs Empathy: What Is the Difference?⏎What Is Empathy?⏎Empathy is the ability to put yourself in someone else’s shoes. In other words, it’s the ability to feel or understand another person. Experts suggest it’s innate in
-
they resemble the pointy nose and ears of a fox.⏎By the way, that likeness inspired another name — fokkusu feisu — which, because it’s written in katakana and includes the English words “fox” and “
-
78 degrees Celsius, even for a few seconds, is enough to kill most bacteria, viruses, or parasites. Toxins produced by bacteria<|endoftext|>Dit artikel staat op icthealth.nl. Link: https://ictandhealth.com/u
-
in a big world. Bravery is having the courage to go out every day and tackle what comes, no matter what comes. These creatively structured poems explore what it means to cultivate respect for self and others, the meaning of good qualities, and
-
19<|endoftext|>毋忘在莒(Pinyin: Wu Wang Zai Ju) is a famous Chinese idiom (成語, Pinyin: Cheng Yu). It is sort of like the Latin idiom left in the European language. Most are
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: when I start a new project, I start with the most recent recent recent recent recent recent recent recent recent recent recent recent recent recent recent recent recent recent recent... (I start writing, but I don're...) 都说都都说都都说
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
varied, from about 5.5 to 8 inches (13<|endoftext|>by Kong Dehao⏎The search for extra-terrestrial intelligence (SETI) celebrates its 58th anniversary this year (Nature, 20
-
had twice been admitted for “exhaustion” and doctors were observing the chronically fatigued young man. Beds of wounded men lined room around him.⏎Everything fell to a disciplined hush when the six-foot, two-inch General George S
-
on account of race or color.⏎With the Civil Rights Act turning 50, the success of the struggle for equal rights for women and minorities<|endoftext|>The No Child Left Behind Act, signed into law by President Bush January 8, 2
-
been caught by the Roman authorities, but others had renounced Christianity in order to stay alive. At first no one had much choice in the matter; too few clergy were willing to make themselves known
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
- L22:309 strongly suppresses
- L20:1423 strongly excites
- L19:1317 strongly suppresses
- L21:468 strongly suppresses
- L23:625 strongly excites
- L23:176 strongly excites
- L23:1210 strongly excites
- L21:2936 strongly excites
- L23:287 strongly excites 'are' after a plural subject
- L19:1422 strongly suppresses
- L20:2453 strongly suppresses blank slot before a date
- L19:545 suppresses
- L18:444 excites
- L21:262 suppresses
- L21:853 suppresses 'can' after 'and' or a noun subject
- L19:1743 suppresses
- L19:1424 suppresses
- L23:1832 excites
- L19:1139 suppresses
- L23:2540 excites
- L19:880 excites
- L23:320 excites
- L22:228 suppresses
- L17:1583 suppresses
Feeds later layers
- L25:586 strongly excites
- L27:321 strongly excites
- L26:467 strongly excites
- L27:461 strongly excites
- L25:347 strongly excites
- L26:1097 strongly excites
- L26:1844 strongly excites
- L25:108 strongly suppresses
- L26:926 strongly excites
- L26:72 strongly suppresses 'they' or 'we' after where or who
- L27:500 excites
- L25:104 excites
- L26:669 excites
- L26:2147 excites
- L26:874 suppresses
- L25:629 excites
- L27:480 excites
- L26:764 excites
- L27:1313 excites
- L27:969 excites
- L27:60 suppresses
- L25:364 suppresses
- L27:1256 excites
- L26:676 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
- has
- $is
- hasn
- knows
- comes
- includes
- requires
- goes
- is
- applies
Suppresses
- 都是
- 都要
- 均
- 都不
- 都无法
- 都
- 都不会
- 都非常
- ayers
- 都可以