'of' after end, top or bottom
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. 5 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 lights up on the word 'of' in place or time phrases like 'end of', 'beginning of', 'top of' and 'bottom of the page'. It is one of the strongest neurons in this set, but turning it up or down did not show a clear theme in the written 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.
-
currently based out of USCG’s Research and Development Centre in New London Connecticut, approximately 375 nautical miles from where the Titanic sank. From beginning of February to the end of July (ice season), the IIP patrols the oceans
-
but there was hardly ever anything to be seen due to bad weather or nighttime ;). The site states that the cam needs to be rehoused as of end of December 2013 which has not happened yet.⏎If the cam comes
-
America's working men and women by setting and enforcing standards, and providing training, education and assistance. For more information, visit www.osha.gov.<|endoftext|>End of the year math review for 2nd grade can be FUN! 2nd
-
at the top of your screen near the toolbars. Then right-click and select AutoCAD. A
-
signed up to receive the notifications of new studies, you can do that today (the form is at the bottom of the page).⏎Esther – Day 3⏎Xerxes wanted to show
-
emissions, driven by driving efficiency measures and lightweighting, is therefore going to increase the absolute share of emissions generated in the production of the car and in its end-of-life phases.⏎Full LCA analyses on all car models⏎Within the next five
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: ... and the 200-word limit. This is a personal and professional story about the current. The topic is about... the user is...... and the user is... I have to you by... the topic... and so much
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
system, attacking other nerve cells, and spreading to other areas of the brain.⏎In diabetic monkeys, the virus was also found in the so-called entorhinal cortex, which also plays an important role in Alzheimer’s disease. If the
-
metal into a die (or mold) to create a specific shape. Once the metal cools, the casting is removed from the mold, and you have your finished shape (more or less).⏎A good example of casting in modern manufacturing is the
-
; ending in a point: as, a peaked beard.⏎- Having a sickly, thin
-
,” Cook added. “Government scientists have found disturbing concentrations of pesticides in some baby foods. And the U.S. Department of Agriculture has found weed killers widespread in finished tap water. Environmentalists have had important successes in forcing pesticides that presented unaccept
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
- L4:742 suppresses
- L4:1721 excites the gap before a decade or century
- L4:925 suppresses
- L4:1712 weakly suppresses
- L4:2114 weakly excites
- L4:2866 weakly excites the word 'average' before a quantity
- L3:1394 weakly suppresses
- L4:1728 weakly suppresses
- L1:1428 weakly suppresses
- L2:2208 weakly excites
- L4:385 weakly excites multiplication and scaling in math
- L4:2374 weakly suppresses mixed: finance terms, strongly negative on systolic
- L0:415 weakly excites
- L3:1252 weakly excites
- L2:2052 weakly suppresses
- L2:2328 weakly suppresses
- L3:2350 weakly excites
- L4:2025 weakly suppresses
- L2:2593 weakly excites
- L3:1352 weakly suppresses
- L0:1338 weakly suppresses
- L4:1902 weakly excites
- L1:1025 weakly suppresses
- L4:2086 weakly suppresses
Feeds later layers
- L16:1542 excites
- L10:2192 excites
- L15:871 excites
- L17:1445 excites
- L9:2391 excites
- L17:2677 weakly excites
- L18:2148 weakly excites
- L21:1408 weakly excites
- L12:1251 weakly excites
- L13:919 weakly excites
- L11:2202 weakly excites
- L12:637 weakly excites
- L7:1941 weakly excites
- L11:350 weakly suppresses
- L17:2052 weakly suppresses
- L6:766 weakly suppresses
- L23:2043 weakly excites
- L17:1138 weakly suppresses
- L10:1765 weakly excites
- L11:1365 weakly suppresses
- L9:3061 weakly excites
- L11:560 weakly excites
- L8:564 weakly suppresses
- L20:1192 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
- acement
- stice
- ñe
- /end
- utoff
- aque
- REATE
- 浒
- Aligned
- ing
Suppresses
- _component
- degrees
- 成分
- 恺
- .persist
- Meteor
- ********************************************************************************
- 生于
- _components
- ippers