unclear: 'no matter' and 'several' before reasons
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. 23 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
The neuron fires on a few different words, mostly "no" in "no matter what" and quantity words like "several" or "few" that come before a list of ways or reasons. It is hard to find one clear theme. Turning it down made the text repeat the word "who" and turning it up gave wandering text.
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.
-
15, 5, 2.5, 2, and 1.8° per step.⏎Step angle = 360°/No of steps per revolution⏎Where m = Number of Phase winding on the
-
the info and together with the real-time checksum authentication which the ZFS file system performs on all drives, you will never have to be concerned about losing any information no matter what.<|endoftext|>also a strong stimulator of creativity.⏎It became clear that
-
product, check<|endoftext|>Help support New Advent and get the full contents of this website as an instant download or CD-ROM. Includes the Catholic Encyclopedia, Church Fathers, Summa, Bible and more all for only $19.99...⏎A
-
Feelings of frustration⏎* Feelings of incompetence⏎* Feelings of loneliness⏎* Feeling of powerlessness⏎* Believing that things will turn out bad no matter what action is taken⏎* Feelings of low self-esteem⏎* Bel
-
when the value rises. Shorting has a high potential for loss, so it’s important to understand the risks.⏎Cryptocurrencies can be shorted in a few different ways, and it’s good for investors to learn how to short them,
-
too far afield; but it is clear from all that is known of the<|endoftext|>Obese & Fat German Shepherd⏎Like humans, dogs can become obese for several reasons, and the effect is just as severe in them. German Shepherds
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: that to have a who-in-the to and to and to and to and to and to and to and to and to and and in and to and and in and to and and in and to and and in and and to and and in
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Eukratides I of Bactria, a land in central Asia (Afghanistan) that was conquered by Alexander. The adoption of the title 'Great King', implying imperial rather than merely royal status, followed some years after Euk
-
several meters wide – floating in the Great Pacific Garbage Patch. These ghost nets are known to be a major hazard to
-
945.”⏎Associate Professor of the Department of Philosophy, Zhanna Simonova, told the students about the contribution of the Kursk Medical Institute to the Great Victory: “In 1941, the first graduation of doctors of
-
compounds have helped to protect the personal health and safety of humans, they create risks and concerns to the environment that are constantly being researched.”⏎An investigative study by the Associated Press in 2008 discovered that a vast array of pharmaceuticals were
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
- L18:1265 suppresses
- L18:602 suppresses
- L4:731 excites
- L17:2348 excites
- L3:914 excites "any" in negated sentences
- L11:233 suppresses
- L4:326 suppresses
- L10:167 weakly suppresses
- L22:2882 weakly suppresses
- L13:2114 weakly suppresses
- L16:1433 weakly excites
- L20:1917 weakly suppresses
- L20:2278 weakly suppresses
- L16:1333 weakly excites 'two' after 'divided into' or 'there are'
- L0:973 weakly excites
- L5:1146 weakly excites
- L18:635 weakly excites
- L21:20 weakly suppresses
- L15:2272 weakly suppresses
- L16:1851 weakly suppresses
- L9:1363 weakly excites ordinal numbers like 16th or 1st
- L6:972 weakly excites "well" after "how"
- L7:232 weakly excites
- L22:404 weakly excites
Feeds later layers
- L24:258 weakly excites
- L24:35 weakly excites
- L26:86 weakly excites
- L24:2538 weakly suppresses
- L24:158 weakly excites
- L25:544 weakly suppresses
- L24:2753 weakly suppresses
- L24:1263 weakly suppresses
- L27:1151 weakly suppresses
- L24:2914 weakly suppresses
- L25:112 weakly suppresses
- L25:1404 weakly suppresses
- L27:440 weakly excites
- L24:1080 weakly suppresses
- L27:1941 weakly excites
- L24:116 weakly excites
- L27:146 weakly excites
- L24:148 weakly suppresses
- L25:1224 weakly excites
- L26:144 weakly suppresses
- L24:111 weakly excites
- L25:3045 weakly suppresses
- L25:1135 weakly excites
- L27:1651 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
- оч
- recent
- 原因之一
- AccessType
- 部分
- 小
- oto
- 东西
- itim
- 主要
Suppresses
- 会觉得
- 在内的
- етесь
- 所说
- 何况
- 所做的
- who
- ersion
- 您好
- saja