period in U.S. and initials
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. 1 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 dot in abbreviations like "U.S." or an initial like "A.S.", where a period is part of a name rather than the end of a sentence. Its negative side shows up when a bracket closes a parenthetical phrase. Turning it up gave text that wandered off topic, so the steering doesn't clearly show what it does.
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.
-
suggested the name “Town of Stinnett” in honor of A.S. Stinnett, and this name was accepted by the Rock Island Railroad. The
-
.,” and 40 percent replied “don’t know.”⏎As the lives of English-language learners change because of attending U.S. schools and interacting with other children in this country, so, too<|endoftext|>There has
-
A notice in the Arkansas newspaper after she died gave the date of the auto accident, which had occurred<|endoftext|>Government issues smoky skies bulletin⏎Smoke from U.S. wildfires is impacting the south coast of B.C., including the Gulf Islands.⏎The
-
OBJECT.⏎SUBTRACTIVE SPACES AND THE DEEP WALL.⏎INDEX OF BUILDINGS.<|endoftext|>"The U.S. Food and Drug Administration announced today that it has approved the influenza vaccine formulation for the
-
more about career and technical education programs of study, talk to your counselor or career and technical education teacher.⏎Career/Technical Education can have positive impact on H.S. graduation rates⏎The chart below shows the impact taking 240 hours of
-
Report Demonstrates U.S. Cotton’s Environmental Progress, Opportunities⏎February 28, 2022⏎The report provides fresh insights for U.S. cotton’s extensive research program as the industry addresses impacts and works toward 20
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 am a 22nd year old, I have a good heart, I have a good ... I am just too curious. I have a good time, I have a good ... I am just too curious. I am not a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
VRMS and IRMS sin φ perpendicular to VRMS as shown in Figure 4.56.⏎(i) The component of current ( I RMS cos φ) which is in phase with the voltage is called active component. The power consumed by
-
structure may be inherited after Recombination. For example, primordial magnetic fields may be responsible for very large scale filaments ( 100 Mpc) as discussed later. Moreover, the existence of periodic structures forming a lattice, actually
-
ilic substitution. It is when an electron-poor reagent (an electrophile, E+) reacts with the electron-rich aromatic ring ( a nucleophile) and substitutes for one of the ring hydrogen. Electrophilic Substitution has
-
( 15 pF or 33 pF ) may be selected.⏎The
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
- L0:178 suppresses
- L0:1301 suppresses
- L0:1224 suppresses
- L0:1354 weakly suppresses
- L0:592 weakly excites
- L0:1 weakly suppresses
- L0:1841 weakly suppresses
- L0:2493 weakly suppresses
- L0:2585 weakly excites
- L0:989 weakly suppresses
- L0:1036 weakly excites
- L0:1825 weakly suppresses
- L0:2920 weakly suppresses
- L0:512 weakly suppresses
- L0:2770 weakly suppresses
- L0:1374 weakly excites
- L0:1626 weakly suppresses
- L0:937 weakly excites
- L0:2308 weakly excites
- L0:2311 weakly excites
- L0:2083 weakly suppresses
- L0:2300 weakly excites
- L0:2752 weakly suppresses
- L0:432 weakly excites
Feeds later layers
- L11:1483 weakly suppresses
- L2:647 weakly suppresses
- L24:1417 weakly suppresses
- L15:2647 weakly excites
- L2:2832 weakly suppresses
- L2:2094 weakly excites
- L19:1635 weakly suppresses
- L14:1305 weakly excites
- L3:1248 weakly excites
- L15:1133 weakly suppresses
- L21:1875 weakly excites
- L13:456 weakly suppresses
- L7:533 weakly suppresses
- L18:2755 weakly excites
- L5:1948 weakly excites
- L26:1192 weakly excites
- L9:2816 weakly suppresses
- L24:2608 weakly excites
- L12:2849 weakly excites '-based' after a place name
- L8:1278 weakly excites 'relevant' before a noun
- L25:1938 weakly excites
- L17:103 weakly suppresses
- L2:346 weakly suppresses
- L23:316 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
- oub
- egasus
- or
- orate
- rel
- arios
- 适应
- 后
- heit
- ис
Suppresses
- za
- <Resource
- dden
- .setEditable
- lda
- //<
- intColor
- newY
- ddb
- ="<