'of' after a title or event noun
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
The neuron fires on small connecting words like 'of' or 'into' that come after a noun or verb and before a name, date or place, such as 'Captain of the Guard' or 'in October of 1904'. It is mixed, and the steering samples don't clearly match.
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.
-
that changes how the⏎cd command works. Normally when you try and cd⏎into a directory, it will only try and change to the path specified from the current
-
ennials and subshrubs) and 2008 (volume 2- ground covers and edgers). Sadly, Carolyn Singer passed away in July of 2019, but her writing still provides vast insight into foothill
-
basic necessities of daily living.⏎Pennsylvania state law required that mining companies and company stores be incorporated separately. Therefore, the Heilwood Company was incorporated in October of 1904 for “the purpose of carrying on a general mercant
-
who have coronary artery disease, especially those who have suffered myocardial infarctions ("heart attacks") in the past⏎- Individuals who have congestive heart failure of any cause⏎- Individuals with intrinsic heart muscle disease such as hypertrophic cardiomy
-
no keys can map, or hash to. So this can be quite a topic.⏎How does hashing relate to hash joins? In hash joins, we will be joining two tables. We read the first table and apply a hash function to each record
-
, as well, such as by coughing into a contracted elbow.⏎Moreover, stay aware of the most recent information on the outbreak
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 calculus of one variable (to use) in the form of a function of a day. When I come to know it, I can use it to be in the country, and the other side of the day. So, if the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
twig among millions on the<|endoftext|>Commemorations for Kiahk 15⏎This day marks the departure of St. Gregory, the Patriarch of the Armenians who was a martyr, without bloodshed (A Confessor). This
-
swung open and the Captain of the Guard strode into the room with a royal warrant for Cromwell’s arrest on
-
1837. A hero of the Civil War and Spanish-American War, he was the first, and to date only, person promoted to Admiral of the Navy.⏎Marihuana Tax Act of 1937⏎On August
-
the next solar maximum, due in 2013, the chance of big magnetic storms is on the increase. On average you might expect to see aurora in the far north of Scotland every few months, but less often as you travel
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
- L6:272 weakly suppresses
- L6:1799 weakly excites
- L5:860 weakly excites
- L4:2022 weakly suppresses
- L2:2420 weakly excites
- L1:1768 weakly excites
- L4:2055 weakly suppresses
- L0:2352 weakly excites
- L6:1200 weakly suppresses
- L6:1350 weakly suppresses
- L5:856 weakly suppresses
- L5:2200 weakly suppresses
- L4:1528 weakly excites
- L3:1314 weakly suppresses
- L1:1583 weakly excites
- L3:728 weakly suppresses
- L5:1560 weakly excites
- L6:2059 weakly suppresses
- L5:2636 weakly suppresses
- L6:1288 weakly excites
- L4:2560 weakly excites
- L1:2109 weakly excites
- L5:2528 weakly excites
- L2:1977 weakly excites
Feeds later layers
- L19:1442 weakly suppresses
- L11:825 weakly excites
- L9:116 weakly suppresses
- L16:1272 weakly excites
- L9:209 weakly suppresses
- L20:2770 weakly suppresses
- L21:1315 weakly excites
- L19:2486 weakly suppresses
- L10:1034 weakly excites
- L18:1428 weakly excites
- L13:743 weakly excites
- L10:930 weakly excites
- L17:801 weakly suppresses
- L9:296 weakly excites
- L9:646 weakly excites
- L8:364 weakly suppresses
- L22:157 weakly excites
- L10:2035 weakly excites
- L15:620 weakly excites
- L16:1984 weakly excites
- L13:2260 weakly excites
- L25:2403 weakly suppresses
- L20:285 weakly suppresses
- L24:1816 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
- cross
- null
- final
- key
- :e
- estone
- prime
- dup
- ing
- NaN
Suppresses
- achable
- ось
- houses
- счита
- 药师
- 级以上
- ��
- classified
- angers
- hausen