'of' in 'in the way of'
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' in set phrases like 'in the way of', 'stand in the way of' and 'in the form of'. Turning it up did not produce text that clearly reflected this phrase.
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.
-
9.9 per cent. Dentists attribute a large part of this to poor flossing habits. Forgetfulness or a busy lifestyle often gets in the way of oral hygiene. But the truth remains that a lack of flossing regimen can
-
health service providers will need to plan for a significant increase in obesity-linked comorbidities.⏎The researchers were particularly interested to know what to expect in the way of co-morbidities, including hypertension, impaired glucose tolerance (which leads to
-
new changes back down on Earth.⏎“A lot of the things we developed to engage in the space race and space exploration eventually filtered down to the public in the way of products and services that we use today and don’t even think about,” added K
-
-In-Chief⏎Eileen Collins, the first female to command a space mission, is a determined down-to-earth woman who never let anything stand in the way of her dream.⏎August 1999: NASA, Texas: Eileen
-
interfere with your body’s chemical signals and prevent you from falling asleep quickly.⏎» Keep kids caffeine-free. While a can of soda may not stand in the way of an adult getting a good night’s sleep, the smaller body of a child can
-
that acknowledges that provider and patient may have different experiences or different goals for treatment.⏎But lack of diversity among providers and poor communication with patients can get in the way of making decisions together. To get around obstacles, the Race and Health Disparities
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 most recent 5000.0001.0001.0001.0001.0001.0001.0001.0001
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ago shoreline.⏎In its 2017 plan, the city's Sustainability Advisory Board identified a need to focus on education and outreach on shoreline restoration, and to encourage private property owners to use incentives like the Storm Water Credit Program, said Elizabeth
-
who remain untreated. Single-cause models of suicide (i.e., social factors or mental illness) do not do justice to the complexity of the processes involved and, therefore, do not provide useful information about the etiology and prevention of suicide in
-
ers, infrared TMDs, and TMDs that do not have a current and valid Certificate of Calibration Testing. These devices can be challenging to read and, because they only show the
-
preserve natural habitats, educate others about their role in conservation and to protect endangered species and environments. Specifically, our program includes courses in forestry, soils
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
- L0:1 weakly suppresses
- L0:1355 weakly excites
- L0:1458 weakly suppresses
- L1:1735 weakly excites
- L1:460 weakly excites
- L1:189 weakly suppresses
- L0:684 weakly suppresses
- L1:2661 weakly excites
- L1:2444 weakly suppresses
- L1:2689 weakly excites
- L0:1279 weakly excites
- L1:24 weakly suppresses
- L1:1540 weakly excites
- L1:186 weakly suppresses
- L0:1075 weakly excites
- L1:718 weakly suppresses
- L1:625 weakly suppresses
- L0:2127 weakly excites
- L0:538 weakly suppresses
- L0:1730 weakly suppresses
- L1:2119 weakly excites
- L0:84 weakly suppresses
- L0:1751 weakly suppresses
- L0:1833 weakly excites
Feeds later layers
- L3:1191 weakly suppresses
- L10:672 weakly suppresses
- L9:835 weakly suppresses
- L8:2175 weakly excites
- L15:211 weakly excites
- L11:432 weakly suppresses
- L23:2992 weakly suppresses unclear: first pieces of split words
- L22:1502 weakly excites
- L24:2148 weakly excites
- L13:1241 weakly excites
- L14:952 weakly suppresses
- L20:1906 weakly suppresses
- L14:1244 weakly suppresses
- L8:2023 weakly suppresses
- L11:1650 weakly suppresses
- L16:2901 weakly suppresses
- L23:2179 weakly suppresses
- L7:956 weakly suppresses
- L22:2208 weakly suppresses
- L5:2906 weakly excites
- L18:1698 weakly suppresses
- L9:522 weakly suppresses
- L4:160 weakly suppresses
- L16:422 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
- itable
- 或者其他
- 텀
- .Gravity
- Deployment
- qué
- .getElementsByName
- arbitrary
- 纬
- обы
Suppresses
- res
- xba
- resc
- [res
- 小时前
- idor
- izr
- nist
- rê
- edo