'to' after attempt or seek verbs
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. 7 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 'to' when it follows verbs like 'sought', 'attempt' or 'seeking', right before the goal of the action is named. Turning it up did not clearly show a theme in the output.
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.
-
constantly sought to maintain a harmonious balance between man and nature. From all accounts of his life
-
their attempt to “tighten the building” through weather-stripping and decreased ventilation keeps air
-
his predecessor and was willing to institute badly needed political and economic reforms that attempted to prevent the country from collapsing into anarchy. These reforms, however, were largely
-
.⏎The split between the two sects is much greater now than it was before that Revolution which sought to bring about Islamist rule across the Arab world with the Shia Ayatollah Khome
-
into Florida, had left it for a time unmolested. The French Protestants, attempting to colonize under Ribaud, built Charlefort at Port Royal in 1
-
: it gives up on gun violence. It sees mass shootings and weapons-related deaths that dwarf every other country. The United States is so saturated with guns that seeking to control them is futile. People own and use guns made in the early 1
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 fact that the two numbers 1, 2, 3, 4, 5, 6 are, say, the two numbers 1, 2, 3, 4, 5, 6,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
schwannomas, meningiomas, angiomas - arteriovenous malformations (AVM), metastases, trigeminal neuralgia and choroidal melanomas.⏎In the body and organs Cyberknife is also used: prostate
-
Find Common Factors⏎How to Simplify Expressions⏎A straight forward approach to simplifying expressions⏎Integer Practice Problems - Addition⏎Review and Practice Problems for adding positive and negative numbers⏎<|endoftext|>In physics, a conservation law states that a particular measurable
-
angle of incidence at large angles and a flatter response to different wavelenghts over the visible. Consequently, graded-index films are desirable in automotive applications where colorless coatings are required. A sol-gel process is presented for the formation of
-
, like diabetes or hypertension⏎Diagnosing retinal disease:⏎Your eye doctor can check for chronic conditions like macular degeneration and diabetic eye changes (retinopathy) by doing a complete eye exam, which includes dilation and examination using special
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
- L4:2577 suppresses
- L5:2777 suppresses
- L6:2758 excites
- L5:1767 suppresses
- L4:1490 suppresses
- L6:1737 weakly suppresses
- L6:268 weakly excites
- L5:1890 weakly excites
- L4:1160 weakly excites
- L4:1866 weakly excites
- L0:935 weakly excites
- L5:924 weakly suppresses
- L6:885 weakly excites
- L0:1101 weakly excites
- L6:670 weakly suppresses
- L5:1293 weakly excites 'of' in organisation names like 'Ministry of'
- L4:1357 weakly suppresses
- L6:990 weakly excites
- L4:2431 weakly excites
- L6:2249 weakly excites
- L5:1399 weakly suppresses
- L4:173 weakly excites
- L6:2899 weakly suppresses
- L5:2564 weakly excites
Feeds later layers
- L12:2370 excites
- L8:2655 excites
- L11:2277 excites
- L8:1465 excites
- L13:1392 excites
- L11:2117 excites
- L10:1010 excites 'in the process of' and continuing actions
- L11:2119 excites
- L25:1335 excites
- L10:1258 excites
- L11:2680 weakly excites
- L19:797 weakly suppresses
- L9:918 weakly suppresses
- L10:340 weakly excites
- L18:1890 weakly excites
- L22:1483 weakly excites
- L12:1989 weakly suppresses
- L21:2304 weakly suppresses
- L10:1311 weakly excites
- L12:1474 weakly excites
- L25:1473 weakly excites
- L12:1875 weakly excites
- L11:3041 weakly suppresses
- L9:640 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
- agements
- agement
- attempt
- тики
- ARGIN
- attempts
- Candidates
- UBLE
- persec
- đuổi
Suppresses
- Soc
- orro
- 总队
- testimon
- far
- вой
- aths
- plings
- plastics
- 廊