'nomad' and 'nosocomial' word stems
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. 23 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 start of the word 'nomad' (nomadic, semi-nomadic) and a few similar pieces like 'nosocomial'. When it is turned up, the model's writing drifts into random, unrelated text about planning, plastics and 3D models.
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.
-
6 × 107.2 cm (52 3/16 × 42 3/16 in.)⏎- Goodman Fund<|endoftext|>Nomads, farmers,
-
the 10th century. The Cossacks were known for their military prowess and self-confidence. The name derives from the Slavic language Kasak “nomad”, “free man”, the term is first mentioned in a Ruthenian
-
animals, nuts and grains eaten on a "catch as you can" basis. Nomadic people learned to render fat from animals and how to use it for cooking and
-
Voortrekkers comprised two groups from the eastern frontier region of the Cape Colony, semi-nomadic pastoralists known as Trekboers, and established farmers and artisans known as
-
moud is part of the line of villages that run north-south, a line that can be seen as marking, traditionally, the end of the semi-nomadic agricultural land and the beginning of the Bedouin land.⏎In this location
-
is a common bird, gregarious and nomadic in response to water levels.⏎It is usually quiet, but does give gut
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 can only 100% 100% 100% 100% 100% 100% 100% 100% 100%
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Most affected individuals die before or soon after birth and they rarely survive beyond the first year of life.⏎Today we assess the risks of Down’s syndrome and other chromosomal abnormalities.
-
attention has been drawn to the potential implications of transgenerational inheritance for human health, so far there is little support."⏎Shutting down a transposon in gametes and the resultant offspring is one thing – it’s a pretty straightforward molecular mechanism
-
age isn’t the only risk factor for male breast cancer. Other possible factors include exposure to radiation, obesity, liver disease and Klinefelter Syndrome, a chromosomal abnormality that affects only males. Generally, the chromosomes that make up male
-
Vs are activated and produced in high quantities during the implantation of the embryo. They are hypothesized to possess immunosuppressive properties, suggesting a role in gestational immune tolerance, thereby protecting the embryo from its mother's immune system. Also
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
- L21:1817 weakly suppresses
- L22:1767 weakly suppresses
- L5:8 weakly excites
- L20:2203 weakly suppresses
- L12:2724 weakly excites
- L17:230 weakly excites
- L19:334 weakly suppresses
- L22:151 weakly suppresses
- L7:2193 weakly excites
- L10:2579 weakly excites
- L19:2193 weakly excites
- L5:364 weakly excites
- L16:1303 weakly excites 'that' starting a clause after a noun
- L15:2958 weakly excites
- L18:2837 weakly excites
- L15:1595 weakly suppresses
- L11:169 weakly excites
- L18:553 weakly excites
- L15:1932 weakly suppresses
- L12:87 weakly excites
- L21:505 weakly suppresses
- L22:2729 weakly excites
- L3:1679 weakly suppresses
- L14:1709 weakly suppresses
Feeds later layers
- L24:2812 weakly suppresses
- L24:909 weakly suppresses
- L24:3037 weakly suppresses
- L24:2327 weakly excites
- L25:2402 weakly suppresses
- L25:1478 weakly excites
- L27:2277 weakly suppresses
- L25:1117 weakly suppresses unclear: names and 're' word starts
- L24:304 weakly excites
- L27:1427 weakly suppresses
- L24:2423 weakly suppresses
- L27:1075 weakly excites
- L26:1228 weakly excites
- L24:1900 weakly excites
- L27:2558 weakly excites
- L24:2250 weakly suppresses
- L27:1820 weakly excites
- L27:1457 weakly suppresses
- L25:1039 weakly excites
- L27:2357 weakly excites
- L24:654 weakly suppresses
- L27:2233 weakly excites
- L24:1429 weakly excites
- L24:1941 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
- ina
- nea
- nargs
- eof
- lover
- �ん
- _EXPORT
- イト
- ctrl
- ival
Suppresses
- aggio
- aggi
- osaurs
- Madden
- 来
- atie
- plastic
- abilité
- pps
- diz