'the' before a noun after '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. 10 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 'the' when it comes right after a preposition like 'of' or 'to', just before a noun, as in 'the back of the book' or 'the stem of the plant'. Its negative side is weak and not clearly related. Turning it up gave text that did not clearly show this pattern.
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.
-
for goods or services in some instances.⏎What are the different types of crypto on the market?⏎To put it simply, there are literally thousands of different cryptocurrencies on the
-
adults focus on the value of being in the class and helping the class become a successful place of learning.⏎The instructor presented a survey to the students on the first day about the merit of participation; and, how participation can
-
behind enemy lines for much longer than this and by 15 November all members of 1st Para Bn's detachment and most of 2nd Para Bn's detachment were unaccounted for.⏎The mission was extremely hazardous for the
-
of symmetry as well).⏎: A parabola is the set of all points in a plane that are equidistant from the focus and the directrix of the parabola. [Adopted from iCoachMath]⏎Focus:Focus
-
organized to allow easier administration of the tests in the classroom. Each test has a total score of 100.⏎An optional listening exercise at the back of the book for each unit, using the recordings from the Student’s Book.⏎An Answer
-
by the plague bacillus, ingests bacilli which may multiply to such an extent that they completely block a valve-like organ at the entrance to the stomach of the flea. When such a flea deserts the host rat and transfers to a human
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: as a point in the past, and in the context of the past, it's important to think of it as a tool that the people of the day, and in the sense, that it's important to be remembered as a historical figure
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
outnumber the colonists. New England, New York, Virginia, and South Carolina were scenes of numerous wars in the 17th century. In 1622, Indians killed a third of the one thousand settlers in Virginia. Another
-
a (in male striped seals) or from the esophagus (in male walruses) and serve to alter the specific gravity of the body; and (4) in the sperm whale, a blind air sac about 1 cu m
-
realish, 15. James Grealish (Michl), 16. John Carr (Pat), 17. Michael Sheridan, 18. Peter Grealish
-
1989, it was among the first 25 featured films to be preserved in the Library of Congress’ National Film Registry, and in 2008, the American Film Institute named it the greatest animated film of all time
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
- L8:149 suppresses
- L9:152 suppresses
- L9:259 suppresses
- L9:749 excites
- L6:93 suppresses
- L8:9 excites
- L6:136 excites
- L7:26 suppresses
- L7:112 suppresses
- L9:682 excites
- L9:103 suppresses "as" or "since" giving a reason
- L8:182 excites
- L8:8 suppresses
- L4:274 excites
- L7:1197 suppresses
- L9:145 suppresses
- L6:205 suppresses
- L8:197 suppresses
- L8:506 suppresses
- L7:35 suppresses
- L8:71 excites
- L6:1006 excites
- L8:179 excites
- L9:270 suppresses
Feeds later layers
- L11:222 suppresses
- L17:334 excites
- L17:29 weakly excites
- L11:234 weakly excites
- L16:2279 weakly excites
- L18:660 weakly excites
- L20:1169 weakly suppresses
- L27:258 weakly excites
- L11:796 weakly suppresses
- L15:764 weakly excites
- L11:56 weakly excites
- L12:77 weakly excites
- L12:1844 weakly suppresses
- L17:43 weakly excites
- L20:1621 weakly suppresses
- L15:487 weakly excites
- L16:1630 weakly excites
- L17:459 weakly excites
- L13:23 weakly excites
- L12:808 weakly excites
- L20:17 weakly excites
- L27:309 weakly excites
- L11:1776 weakly excites
- L19:398 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
- ario
- eres
- 游艇
- wood
- riba
- wend
- 下面是
- rowning
- edian
- este
Suppresses
- too
- syst
- .'.$
- ’)
- ’:
- .Seek
- shrimp
- Hun
- .')
- gu