"on" in "went on to"
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 word "on" in phrases like "went on to" or "going on", where a verb is followed by "on" and another verb. Turning it up did not produce a clear change in what the model wrote.
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.
-
last week when climate advisors in the U.K. warned that the younger generation will be the ones paying the price of climate change. The Committee on Climate Change went on to say they’re “acutely concerned” over the U.K.’s lack
-
were elected, or 39%. The success rate for plurality winners in U.S. Senate races is much worse, as only four of 35 went on to win (11%). Candidates who went through a primary runoff tended to fare
-
frying pan?⏎How Is the Process of Coffee Beans Roasting Going On?⏎The process of roasting is required to bring coffee beans to the condition suitable
-
to Alzheimer’s in roughly 10% to 15% of cases each year, according to the study.⏎Forty-five percent of the participants went on to develop serious dementia consistent with Alzheimer’s disease. Those who had high cholesterol,
-
all American citizens.⏎Houston, tragically, would die before his strategy came to fruition in the Brown suit, but Marshall would argue the case victoriously and go on to become the first African-American Supreme Court justice—always crediting his mentor for
-
’ should be adhered to. Yet, the current amendment goes on to give colossal discretion to the forest bureaucracy in the name of curbing colonial power
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 now a student of the University of the Science of the Future. So, let me know. I am also a student of the University of the Science of the Future. I am also a student of the University of the Science of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
o correction factor, allow mathematical calculation of the affects of air density on the wide-open-throttle horsepower and torque. The dyno correction factor is simply the mathematical reciprocal of the relative horsepower value.⏎Originally, all of the major US auto manufacturers were
-
perpendicular lines are the lines that intersect at 90 degrees.⏎Answer: Lines with the same slope are parallel and if the slope of one line is the negative reciprocal of the second line, then they are perpendicular.⏎Let's see how to identify
-
measure of how easily electricity flows along a certain path through an electrical element. The SI derived unit of conductance is the siemens (formerly referred to as the reciprocal ohm or mho). Oliver Heaviside coined the term in September
-
such as polymer melt.⏎Research in the Polymers and Industrial Mathematics group focuses on the mechanics of polymers and other complex fluids, free-surface flows and inverse problems. We are also concerned with the development and implementation of novel numerical and computational
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
- L0:2553 weakly excites
- L1:2882 weakly suppresses
- L2:2974 weakly suppresses
- L2:2913 weakly excites
- L2:111 weakly suppresses
- L1:1246 weakly suppresses "word" after a language origin phrase
- L0:178 weakly suppresses
- L1:146 weakly excites
- L0:1445 weakly excites
- L2:1424 weakly excites
- L2:1710 weakly suppresses
- L2:2426 weakly excites
- L1:951 weakly excites
- L0:1224 weakly suppresses
- L2:66 weakly excites
- L2:2125 weakly excites
- L2:1494 weakly suppresses
- L1:2653 weakly excites
- L1:1464 weakly excites
- L1:2040 weakly excites
- L0:1337 weakly excites
- L2:2658 weakly excites
- L2:2326 weakly suppresses gravity word pieces (gravit-)
- L2:366 weakly suppresses
Feeds later layers
- L10:940 suppresses
- L9:946 suppresses
- L15:423 excites
- L4:303 weakly excites
- L24:2008 weakly suppresses
- L10:2453 weakly suppresses 'of' after publisher or division nouns
- L9:1789 weakly suppresses
- L4:2864 weakly suppresses
- L11:1724 weakly suppresses
- L5:2582 weakly suppresses
- L13:949 weakly suppresses
- L21:1974 weakly excites
- L12:552 weakly suppresses
- L10:2020 weakly suppresses
- L6:1262 weakly excites
- L15:808 weakly excites
- L19:954 weakly excites
- L13:311 weakly excites
- L10:1522 weakly excites
- L6:1692 weakly excites
- L15:2238 weakly excites
- L9:2389 weakly suppresses
- L13:525 weakly suppresses second half of an opposite pair
- L17:1951 weakly suppresses
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
- ourn
- NR
- iol
- .IContainer
- iom
- 设计师
- NavController
- nr
- esign
- nghiệm
Suppresses
- dır
- alık
- itez
- .bat
- backs
- сь
- .backward
- ếu
- inspace
- oğlu