modifier words before their head word
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. 14 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 fires on a word that qualifies the word coming right after it, such as 'strongly' in 'strongly disagree' or 'very' in 'very high'. It is a loose pattern, and turning it up gave rambling or repetitive text rather than anything on 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.
-
classroom environment to avoid having students draw comparisons between the characters in the film and their classmates.⏎Before Watching: Consider whether you strongly agree, agree, disagree, or strongly disagree with this statement: Sometimes we have to change who we truly are
-
board, each with one of the following statements written in large letters: “Strongly Agree,” “Agree Somewhat,” “Disagree Somewhat,” “Strongly Disagree”⏎Activities / Procedures:⏎1. WARM-UP/DO
-
the tau (τ−⏎) and the tau neutrino (ντ⏎). Each type of neutrino is associated with an antimatter particle, called an antineutrino⏎, which also has neutral electric charge and 1/2
-
food with either high or very high salt levels, which are comparable to a human diet, for eight weeks.
-
than a split second decision to specify materials with recycled or recyclable content. A truly sustainable strategy tries to understand the impact of building, or of not building, and to address the environmental, ecological, cultural, economic, and financial
-
states:⏎|Bit 1||Bit 0||Data Size|⏎The memory system must be able to handle word, half word and byte writes.⏎Memory systems supporting only word writes will have severe difficulties supporting
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 question, to answer? and the two parties? to the question, is the sentence in the the... and... and then, in the first paragraph, the author is? A) the the next the is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
.⏎Sacrificial Anodic Coatings⏎Anodic coatings for the protection of iron and steel substrates are almost totally limited to zinc and aluminium coatings or their alloys. Where coatings anodic to the substrate are applied, the corrosion protection is referred
-
think this problem is relevant for Professor Finnegan's class at WSU.⏎What textbook is this problem found in?⏎Our data indicates that this problem or a close variation was asked in Chemistry: An Atoms First Approach - Zumdahl Atoms
-
(FANC), as well as the p53 signaling pathway. B. In response to intrastrand crosslinks (crosslink), ATR and its interacting partner ATRIP become activated and phosphorylated leading to the induction of several downstream pathways
-
contain chlorine. The EPA allows a certain measurement of chlorine to be in the water and still be deemed “safe”.⏎Chlorine or Chloramines (mix of Ammonia and Chlorine that large cities use) is present as a disin
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
- L11:3040 excites
- L13:1762 excites
- L12:1493 excites
- L12:2861 suppresses
- L13:2503 excites
- L9:946 suppresses
- L12:1897 suppresses
- L10:3054 suppresses
- L13:2563 excites
- L8:2956 excites
- L13:2902 suppresses
- L12:2221 excites
- L13:2435 suppresses
- L12:241 excites
- L11:1420 suppresses
- L11:1035 excites
- L13:1666 suppresses
- L7:753 suppresses
- L9:1436 excites
- L9:750 weakly excites
- L13:71 weakly suppresses
- L11:1770 weakly excites
- L9:3065 weakly excites
- L11:1623 weakly excites
Feeds later layers
- L16:2789 excites
- L15:1063 suppresses
- L15:1380 excites
- L17:1679 excites
- L15:3069 excites
- L17:662 excites
- L18:1373 excites
- L15:1855 excites
- L15:1498 weakly excites
- L18:1314 weakly suppresses
- L17:1133 weakly excites
- L16:561 weakly excites
- L15:187 weakly excites
- L15:1928 weakly excites
- L15:1208 weakly suppresses
- L17:3039 weakly suppresses
- L18:944 weakly suppresses
- L21:2528 weakly suppresses
- L19:1410 weakly excites
- L15:1746 weakly excites
- L15:1719 weakly excites
- L16:2086 weakly excites
- L15:1494 weakly suppresses
- L20:1157 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
- azers
- 删除成功
- 日讯
- 屋里
- 一个重要
- aires
- orphism
- 这几个
- 免责声明
- OperationException
Suppresses
- zas
- nees
- olum
- adia
- remains
- CTX
- ILE
- ��
- footing
- nee