'of' after 'those' or 'that' in comparisons
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 reacts to the word 'of' when it comes right after 'that' or 'those' in a comparison, as in 'different from those of the national sample'. Turning it up did not produce any obvious change in topic or style.
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.
-
adjacent districts of the northern lowlands. Smaller agglomerations of this group are also found along the eastern littoral where their settlements are juxtaposed with those of the Muslims.⏎The Sri Lankan Muslims trace their origin back to Arab traders of
-
ulating layers and one or more copper layers that contain the signal traces and the powers and grounds; the design of the layout of PCBs can be compared to that of the electrical circuit.⏎- Single Layer PCB –
-
.⏎San Antonio has a unique and strategic location. It is on the edge of the Great Plains. The topography, rainfall and water supply were different from that of the East. The Easterners had a forest background. The traditions, laws,
-
G. maedai sp. nov. (middle to ?upper Cenomanian). The habitats of Goshoraia have been extensively compared with those of common Cretaceous, nonsiphonate burrowers, such
-
Indian students reported significantly higher lifetime and last 30-day substance use rates compared with the national sample. Lifetime use rates for American Indian students were higher than those of the national sample for all substance measures except tranquilizers and amphetamines (all
-
, right? See it in action below, and check out the project’s elaborate overview here.<|endoftext|>Our system of government is different from those of all other nations, because part of the political power is vested in the State,
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: "I need to know what the number of the number of the number of the number of the number of the number of the number of the number of the number of the number...? -4444444444
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of Ireland 1911⏎Household Return (Form A) for house No. 15⏎In 1911, James Grealish (
-
molecules with complementary rather than identical structures in juxtaposition. We see here a direct connection to molecular packing in molecular crystals. Aleksandr Isaakovich Kitaigorodsky (left) published a brief paper in 1945, “
-
William Sutherland was one of several Scots who explored the country in the 19th century. He ended a wandering life (he had earlier fought under Garibaldi in Italy) when he settled at Milford Sound in
-
and signed by James. It was witnessed as a truthful return and collected by Constable Thomas Fitzgerald on April 11th.⏎10. James Michael Grealish, 11. James (Jas) Grealish,
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
- L1:1055 weakly excites
- L1:1857 weakly suppresses
- L1:933 weakly excites
- L1:2413 weakly suppresses
- L1:92 weakly excites
- L0:3039 weakly excites
- L1:74 weakly excites
- L0:2951 weakly excites
- L1:10 weakly excites
- L1:912 weakly suppresses
- L1:2242 weakly suppresses
- L0:501 weakly suppresses
- L1:2449 weakly excites
- L0:2836 weakly suppresses
- L1:1541 weakly suppresses
- L1:2822 weakly suppresses
- L0:2523 weakly excites
- L1:303 weakly suppresses
- L0:1951 weakly suppresses
- L0:1042 weakly suppresses "This" starting a labelled sentence
- L1:892 weakly excites
- L0:1271 weakly excites
- L1:2825 weakly excites
- L1:338 weakly excites
Feeds later layers
- L18:2664 suppresses
- L6:1001 weakly suppresses
- L15:275 weakly excites
- L14:173 weakly excites
- L14:2673 weakly excites
- L11:2817 weakly excites
- L15:2608 weakly suppresses
- L13:801 weakly excites
- L26:1228 weakly excites
- L24:1362 weakly excites
- L14:2845 weakly suppresses
- L14:2983 weakly suppresses
- L16:320 weakly excites
- L27:796 weakly excites
- L5:361 weakly excites
- L3:2605 weakly excites
- L19:1475 weakly suppresses
- L3:2372 weakly suppresses
- L9:2262 weakly suppresses
- L6:1722 weakly excites
- L15:1107 weakly suppresses
- L26:2424 weakly suppresses
- L23:1469 weakly excites
- L21:883 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
- ycop
- ahir
- ʏ
- 原著
- ès
- 导入
- ro
- Ts
- 대로
- ays
Suppresses
- politic
- lich
- Mailer
- lix
- plode
- $
- dzi
- hibit
- 当たり
- .false