"of" in time-span phrases
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 phrases about a span of time or extent, like "period of time", "a long period of" or "single layer of". The negative side fires on odd hyphen pieces in numbers and compound words. Turning it up did not produce a clear change in the writing.
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.
-
fatty acids attract each other to the inside of the layer and the hydrophilic ends are on the outside of the membrane in the water, creating a double layer of fatty acids⏎How are bacterial species identified?⏎Cell and colony morphology, chemical composition
-
deep waters off the coasts of<|endoftext|>Antireflective coatings consisting of a single layer of glass with a graded variation in refractive index offer advantages over multi-layer coatings.
-
do have limitations. Cell phones have only been around for about 20 years, so the studies haven’t been able to follow people over a really long period of time. Also, there haven’t been any good studies looking at the effect of
-
26th Illinois Volunteers regiment and was commissioned as Captain on August 10, 1861 by Governor Richard Yates. After less than a month of service, Tinkham was promoted to Lieutenant Colonel on August 31,
-
the employment landscape of the industry. James Cook University professor Ian Atkinson said it was likely some jobs would be lost but expected it to happen over a long period of time.⏎“It’s not just taking workers out and putting a machine in, it
-
and seasons⏎In the task sheet 2, students are reminded that 'climate' refers to the average, or typical, weather conditions observed over a long period of time for a given area. It is determined by factors such as latitude (amount
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 idea that, if you have a time series that is in the format of the United States, and you have to take into account of the number of people, and the amount of work, and the level of efficiency, and the number
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
along a 100-kilometer long north-south linear vent system over the past 5 million years
-
and wheat acreage in Georgia is dropping.⏎To boost the state’s wheat industry and help producers get more out of their crop, University of Georgia Cooperative Extension small-grains specialist Reagan Noland is researching a dual-use system
-
Beach and Broward counties. The C-51 Basin in western Palm Beach County experienced a 1-in-100-year rainfall event with nearly 15 inches of rain over
-
that can detect biomolecules of interest without requiring the additional use of an optical label. They used a tightly focused laser to drill holes through a 100-nanometer-thick layer of gold deposited on a 100-nan
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
- L1:861 weakly suppresses
- L1:712 weakly excites
- L1:231 weakly suppresses
- L1:951 weakly excites
- L1:2126 weakly excites
- L0:1182 weakly excites
- L1:1322 weakly excites
- L0:2705 weakly excites
- L1:1152 weakly suppresses
- L1:1096 weakly suppresses
- L1:1417 weakly suppresses
- L0:2243 weakly excites
- L0:1836 weakly suppresses
- L0:0 weakly excites
- L1:262 weakly excites
- L1:1244 weakly suppresses "of" after "board" or "off"
- L0:2179 weakly excites
- L1:468 weakly excites
- L0:5 weakly excites
- L1:74 weakly excites
- L1:898 weakly excites
- L0:2914 weakly suppresses
- L1:1126 weakly suppresses
- L0:2854 weakly suppresses
Feeds later layers
- L6:1195 suppresses
- L12:575 weakly suppresses
- L3:750 weakly excites
- L19:2369 weakly suppresses
- L9:1163 weakly suppresses
- L14:298 weakly excites
- L8:386 weakly suppresses
- L6:1063 weakly suppresses
- L9:1165 weakly suppresses
- L16:2153 weakly suppresses
- L25:67 weakly excites
- L13:1365 weakly suppresses
- L7:1313 weakly suppresses
- L9:2838 weakly excites
- L11:469 weakly suppresses
- L15:1549 weakly suppresses
- L6:350 weakly suppresses
- L3:492 weakly excites
- L8:359 weakly suppresses
- L3:195 weakly excites
- L7:2159 weakly suppresses
- L15:2308 weakly excites
- L12:696 weakly suppresses
- L20:1611 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
- ienne
- UCH
- ellular
- <Token
- уча
- hiệp
- ATTERN
- atty
- ично
- #:
Suppresses
- الرسمي
- ofs
- CAM
- -pagination
- abcdefgh
- 发展潜力
- obar
- anner
- 不一样
- emple