'of' after 'pattern' or 'patterns'
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. 2 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 'of' when it follows the word 'pattern' or 'patterns', as in 'a pattern of behaviour'. Turning it up did not produce a visible change 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.
-
Washington (1994) contends that Tarquin, understood to be the poem's villain, serves to emphasize a complex pattern of meaning at work in The Rape of Lucrece. Washington maintains that Lucre
-
any element with a dark current, so to speak, will consistently report an exposure value.⏎In fact, with a long enough exposure to darkness, a disturbing pattern of exposure will appear from any CCD, because CCD elements tend to leak current. They
-
how information makes people perceive the world. Authors in many fields have noted patterns of communication that create distrust and enforce dependency by emphasizing danger from external, uncontrollable forces
-
considering both neural and molecular levels. At the neural level we can think of all of these concepts that people apply to themselves and others as patterns of firing in groups of neurons. A sufficiently complex account of neural representations can explain how
-
encian Christians. Indeed, it is perhaps more accurate to view the Mudejars as forming a distinct subsociety separated from the larger Christian society by patterns of intragroup social and cultural behavior, but intersecting with that Christian society in
-
sophisticated level. Instead of building birds' nests, we build sky scrapers. Our minds arrange events in patterns which we call designs. This memo reflects a pattern of events arranged by intent.⏎are Causes of other Patterns of Events? Patterns of events
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 problem is a classic example of a non-integer value, which is a problem that's not a real number value. However, the problem is not a real number value because the user is thinking about it, so I need to check
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
must consider the variety of his receivers.<|endoftext|>Money is what you pay in order to get
-
.⏎Since knowledge, thinking, and rational action are properties of the individual, since the choice to exercise his rational faculty or not depends on the
-
unexpected outcomes, such as sudden drops in temperature in certain regions. This interaction exemplifies the complexity of atmospheric dynamics and its unpredictable impact on local weather conditions. And that is what we
-
1976. One hundred twenty-four years after it was first proposed, Kenneth Appel and Wolfgang Haken cleverly reduced the problem to checking the properties of 1,936 maps by computer. The result was hundreds of pages
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
- L0:1079 weakly suppresses "up" after "open"
- L1:853 weakly excites
- L1:187 weakly suppresses
- L1:296 weakly suppresses
- L0:2965 weakly excites
- L1:712 weakly excites
- L0:187 weakly suppresses
- L1:544 weakly excites
- L1:2893 weakly excites
- L1:2380 weakly suppresses
- L1:1805 weakly suppresses
- L1:1975 weakly excites
- L0:1583 weakly suppresses
- L1:843 weakly excites
- L0:1373 weakly suppresses
- L1:2440 weakly suppresses
- L0:2702 weakly excites
- L1:128 weakly excites
- L1:1273 weakly excites
- L0:2441 weakly suppresses
- L1:719 weakly suppresses
- L1:1571 weakly excites
- L1:1621 weakly excites
- L1:2546 weakly excites
Feeds later layers
- L3:222 suppresses
- L6:607 weakly suppresses
- L4:266 weakly suppresses
- L17:2672 weakly suppresses
- L20:1137 weakly suppresses
- L9:248 weakly suppresses
- L8:1151 weakly suppresses
- L12:1280 weakly excites
- L8:293 weakly excites
- L13:989 weakly suppresses
- L14:1158 weakly excites 'to' starting a purpose clause
- L12:2926 weakly suppresses
- L3:659 weakly excites
- L5:707 weakly excites
- L8:2883 weakly excites
- L11:2774 weakly suppresses 'least' in 'at least' hedges
- L27:2390 weakly excites
- L22:1077 weakly excites
- L7:1757 weakly suppresses
- L16:2461 weakly excites
- L8:1954 weakly suppresses
- L7:478 weakly suppresses
- L9:2784 weakly suppresses unclear: deity names strongly negative
- L6:1015 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
- okit
- ativos
- 养
- отов
- apa
- OVE
- APA
- onna
- หมาย
- jual
Suppresses
- aminer
- grantResults
- ступ
- �
- произ
- honored
- {}".
- غم
- flix
- false