comma inside dates and years
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. 3 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 comma inside a citation or date, such as the comma after a year in "(2014, July)" or "(1990, 1992)". Its strongest negative cases are the word "to" in number ranges like "from 1792 to today" or "1210 to 1274". Turning it up did not give a clear theme in the steered text.
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.
-
M., Chan, K. S.-C., Chan, D. W.-O., & Chung, K. K.-H. (2014, July). Effectiveness of a tiered intervention model in Chinese: A growth model
-
the module.⏎See the Discussion Board Guidelines and Rubrics⏎Download Discussion Board Guidelines and Rubrics.⏎Reh, F. J. (2018, November 01). Here Is a Look at the Role and Responsibilities of Today
-
the extent and patterns of seasonality in FC levels in coastal waters. Water quality monitoring data for shellfish management collected during El Niño (1990, 1992, 1997), and La
-
of having an opportunity to prosper in life.⏎The following two case studies are based on the book The Life Span Human Development for Helping Professionals (2006, 2nd ed. ) by Patriia C.⏎John Winthrops City
-
Probes of Bulk and Surface Compositions of Size-Resolved Sea Spray Aerosol Particles,” Phys. Chem. Chem. Phys., 2013, 15, 6206-6214.⏎Atomic
-
ideal supplement to high school and college-level Holocaust studies, and adult interfaith discussion groups.⏎A Time to Remember⏎USA, 1992, 52 minutes, color⏎Produced by Sy Rotter and Andrzej
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: 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
House? Learn about this and other presidential "firsts."⏎Read more >⏎History of the White House and Washington, D.C. from 1792 to today.⏎Who was the first U.S. president to be photographed?⏎Grades
-
ization.⏎HG is a more severe form of nausea and vomiting of pregnancy (NVP), also called morning sickness. Morning sickness affects anywhere between 50% to 90% of pregnant women. HG is estimated to occur in 0
-
9. Coutances houses a well-known botanical garden and an art museum.⏎Coutances Cathedral is a Gothic Roman Catholic cathedral constructed from 1210 to 1274 in the town of Coutances, Normandy, France
-
a Temple choir and a Board of Relief established to care for Jewish burials and illnesses. Rabbi Jacob Block served B’nai Israel from 1872 to 1880.⏎Disagreements about religious observances and liturgy
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
- L2:1710 weakly suppresses
- L2:84 weakly excites
- L0:2445 weakly excites
- L1:1244 weakly suppresses "of" after "board" or "off"
- L1:541 weakly excites
- L0:1224 weakly suppresses
- L0:901 weakly suppresses 'this' after sentence-initial 'In'
- L2:1905 weakly excites
- L2:2098 weakly suppresses
- L2:1036 weakly excites
- L0:3017 weakly suppresses
- L1:2477 weakly suppresses
- L1:337 weakly excites
- L0:178 weakly suppresses
- L2:872 weakly excites
- L2:298 weakly excites
- L2:2152 weakly excites
- L2:1494 weakly suppresses
- L1:1016 weakly suppresses
- L0:1301 weakly suppresses
- L0:2020 weakly excites
- L2:2263 weakly suppresses
- L1:767 weakly suppresses
- L2:1269 weakly excites
Feeds later layers
- L7:42 suppresses
- L4:546 suppresses
- L8:562 suppresses
- L8:376 suppresses
- L13:2325 weakly excites
- L9:212 weakly excites
- L19:1017 weakly suppresses
- L12:1851 weakly suppresses hyphen inside dates
- L23:2583 weakly excites
- L18:1829 weakly excites
- L7:1014 weakly suppresses
- L7:836 weakly suppresses
- L16:1782 weakly suppresses
- L6:2170 weakly suppresses
- L16:2747 weakly excites
- L6:1702 weakly suppresses "not" or "instead" contrasting a correction
- L5:51 weakly suppresses
- L17:445 weakly suppresses
- L10:2046 weakly excites
- L5:2657 weakly excites
- L16:520 weakly excites
- L5:2118 weakly excites
- L13:2356 weakly suppresses
- L16:1015 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
- ichever
- 共产党员
- 让他们
- ATERIAL
- 每个月
- adesh
- 每一个
- 례
- 洒
- ALAR
Suppresses
- Others
- thur
- ых
- other
- other
- _under
- loaf
- ổi
- outros
- лез