'of' after a measured quantity
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. 6 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 a measurable quantity or setting, like 'a carbon price of $50', 'altitude of 8,000 feet' or 'a value of zero'. Turned up, it pushed the writing toward numbers and amounts. It goes quiet or negative on 'was' or 'is' in comparisons like 'a response time that was'.
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.
-
this month, may result in more legislation, or a timetable for the introduction of further curbs on emissions, for example an international carbon price. A carbon price of $50 per tonne emitted is already accounted for by some companies, C
-
a cabin altitude of 8,000 feet. Many post-abdominal surgery patients have
-
must be visible for the character to be visible.⏎The width and height of the rectangular area are determined by the⏎value of this property.⏎For example a value of zero represents a point in the center. A⏎value of a half represents a
-
a force of enlightenment and it is not only verses of the Quran that testy to that fact
-
majors. Use ANOVA – single factor in Excel’s Data Analysis ToolPak to test whether there is a relationship between the two variables. Use an alpha of 10% Show your hypotheses, ANOVA output, and decision –
-
,190 km2 (459 sq mi) with a basal circumference of 140 km. This makes it by far the largest of the three
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 question is, which of the following is the best? The structure of the question is such that the first data set can be partitioned in a way that the first data set is a group of things. The question is, which of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
bound to the antibody attached to the gold film lining the array holes. When compared with established SPR-based measurement techniques, the flow-through device had a response time that was sixfold faster while measuring PAX8 present at concentrations in the attomolar
-
the indoor air quality of your home or office has an air pollution level that is 2-5 times higher than the level of the air outside its doors.⏎
-
along with a brief explanation written by a professional astronomer.⏎2000 April 21⏎Explanation: Star formation occurs at a faster pace in M82 -- a galaxy with about 10 times the rate of massive star birth
-
Earth. M81 is a spiral galaxy about 12 million light years away that is both relatively
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
- L4:200 weakly excites
- L4:1824 weakly suppresses
- L4:2919 weakly excites
- L5:985 weakly excites
- L1:1399 weakly suppresses
- L3:408 weakly suppresses
- L5:1544 weakly suppresses
- L5:75 weakly excites
- L5:221 weakly excites
- L1:2748 weakly excites
- L5:117 weakly excites
- L3:59 weakly excites
- L5:102 weakly excites
- L3:856 weakly excites
- L2:483 weakly suppresses
- L1:2340 weakly suppresses
- L2:1899 weakly excites
- L3:833 weakly suppresses
- L1:146 weakly excites
- L5:1016 weakly excites
- L3:1076 weakly excites
- L3:105 weakly suppresses
- L5:222 weakly suppresses
- L4:350 weakly suppresses
Feeds later layers
- L7:582 strongly excites
- L10:868 excites
- L12:1467 excites numbers after dates and figures
- L8:229 excites
- L18:1259 excites
- L7:1524 excites
- L8:190 excites
- L7:79 excites
- L19:248 suppresses
- L22:481 excites
- L13:2034 excites
- L19:2450 excites
- L14:827 excites open bracket after a value or citation
- L14:1378 excites
- L14:471 excites
- L12:305 excites
- L24:722 excites 'about' or 'approximately' before a number
- L7:1742 excites
- L18:585 excites
- L20:780 excites 'of' before a measured value
- L23:198 excites
- L11:579 excites
- L16:1169 excites
- L13:287 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
- 甯
- せい
- 和个人
- $log
- .Writer
- 客户提供
- ением
- igion
- PHPExcel
- ồ
Suppresses
- trag
- wer
- ebb
- uilt
- ący
- aket
- ster
- reak
- нст
- Zug