the digit 1 in numbers
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. 22 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 digit '1' when it starts a number, such as the first digit of 140/90 or 1230 degrees. It is a number-position detector, and the steering results do not show a clear effect of turning it up or down.
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.
-
This can be caused by atherosclerosis or narrowing of the blood flow in the heart. ...Read moreSee 1 more doctor answer⏎My BP is 140/120 sometimes, other times 130/9
-
Porcelain is rolled to create the tile, each letter is hand imprinted into the clay, which is then fired in the potter’s kiln at 1230 degrees celsius, making it solid to withstand all elements. The
-
finally evaporating at 212 degrees, making the cookie stiffer, but leaving behind empty spaces that makes it lighter and flaky.⏎- At 144 degrees, the proteins in eggs—which look like tightly coiled springs—
-
.⏎If your blood pressure reading is high, you will probably be asked to come back for repeat blood pressure checks. If you have three visits with readings over 140/90 mmHG, you will be diagnosed with high blood pressure
-
1185 CE), Japanese art flourished, and new forms of art, such as screen painting, began to emerge. The Kamakura period (1185 CE to 1333 CE) saw the rise of
-
beginning of the research, 15 subjects were randomly assigned to either a moist-heat or stretching treatment. The moist-heat therapy involved placing a hot (160-degree Fahrenheit), moist pack in a terry-cloth cover on each
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: what if the first two degrees of pi, pi plus 3, pi plus 6, pi plus 9, pi plus 12, pi plus 15, and pi plus 18, plus the degrees of pi
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
standards will be piloted in Botswana, Egypt, Ethiopia, Kenya and Paraguay. The initiative contributes to WHO’s 13th General Program of Work target of ‘1 billion lives made healthier’ by
-
Wall Street” protests were against the far-reaching negative consequences of neo-liberal capitalism.⏎In the wake of the global financial meltdown, the notion of Capitalism 4.0 has suddenly made headlines and the Davos Forum early this year also addressed
-
– but the story remains. Products of short-lived radioactive isotopes found in some meteorites contradict the 4.5 billion year story. The elements that would have formed primordially in
-
Lorenz developed the concept that tiny effects in one part of a system can lead to big changes in another part. The term butterfly effect first appeared in his 1972 paper Predictability: Does the flap of a butterfly's wings in
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
- L19:418 excites
- L19:723 excites
- L19:2533 excites
- L16:1333 excites 'two' after 'divided into' or 'there are'
- L9:1348 suppresses
- L19:720 excites
- L17:512 suppresses
- L4:205 excites the digit 2 inside citations
- L21:1863 suppresses
- L19:851 weakly excites comma inside numbers with thousands
- L21:2842 weakly excites
- L4:464 weakly excites
- L19:1047 weakly excites
- L16:1012 weakly excites 'one' before 'of' in 'is one of the'
- L19:25 weakly excites
- L12:953 weakly excites
- L21:518 weakly excites
- L9:13 weakly excites
- L0:17 weakly excites
- L9:1032 weakly suppresses
- L17:157 weakly excites
- L5:2867 weakly excites
- L20:416 weakly excites the "1" in years 191x
- L3:659 weakly suppresses
Feeds later layers
- L24:2097 weakly suppresses
- L27:1527 weakly excites
- L24:23 weakly suppresses
- L23:2634 weakly suppresses
- L23:2028 weakly excites
- L24:2720 weakly excites
- L23:2790 weakly excites
- L23:2351 weakly excites
- L23:735 weakly excites
- L25:2991 weakly excites
- L24:1750 weakly excites
- L23:1109 weakly excites
- L25:786 weakly excites
- L25:2908 weakly suppresses
- L24:2632 weakly excites
- L23:284 weakly excites
- L23:2505 weakly excites
- L27:1598 weakly excites
- L23:1563 weakly excites
- L24:2384 weakly excites
- L23:599 weakly excites
- L25:1278 weakly excites
- L25:457 weakly suppresses
- L25:2737 weakly excites 'a' before 'few' or 'couple'
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
- slashes
- uries
- trusts
- UIStoryboardSegue
- ifetime
- chron
- anchor
- 阄
- 邗
- Disqus
Suppresses
- eld
- 馅
- ffield
- °C
- .CommandType
- iative
- ICLES
- ices
- ��
- elden