letter c inside a split word
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. 27 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 lights up on the letter 'c' when it sits in the middle of a word that has been chopped into pieces, like 'vis-c-eral' or 'circum-c-isions'. Turning it up made the model's text fall into repeated word loops, which tells us little about what the neuron 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.
-
chronic pelvic pain are as follows-⏎- Central and peripheral nervous systems: Nerve damage of serious nature may occur due to anomalies in visceral functions and perceptions (visceral hyperalgesia
-
to traffic in 1880. It connected the San Miguel district to Malate and Ermita.⏎The bridge also served Isla de Convalescencia, the island in the Pasig River where the orphanage of Hospicio
-
about the Cassini-Huygens mission visit http://saturn.jpl.nasa.gov/home/index.cfm. The Cassini imaging team homepage is at http://ciclops.org.<|endoftext|>Archive for the ‘transcendental geometry’ tag
-
be an open⏎front protected by glass or a closed front faced with bronze,⏎marble, or granite. (An arrangement of niches is called a⏎c
-
Rights Reserved.|<|endoftext|>Mount Cleveland is on alert level yellow. The french blog Volcanic News Erutpion states it as being restless. http://volcansvanuatueruptionsgb.blogspot.fr/2014/0
-
we can get involved in the fight against cancer.⏎World Cancer Day -- make a difference to the fight against cancer.⏎Discover how to get engaged:http://www.worldcancerday.org<|endoftext|>There are both normal and excessive amounts of anxiety you may experience
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: whyentheententententententententententententententententententententententententententententententententententententententententententententent
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Barriers to Addiction Treatment?⏎Barriers to finding addiction treatment may vary by personal situation, but the most common barriers include:⏎- Geographic Location⏎- Co-Occurring Disorder Treatment Availability⏎
-
ی دکتری معماری منظر، دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران، ایران.⏎عنوان
-
⏎1 دانشیار معماری، دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران، ایران
-
codilian species may also harbor the influenza A virus. Here we analyzed 37 captive crocodilians from two locations in Florida (plus 5 wild bird fecal samples from their habitat) to detect the presence of influenza A virus. Several
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
- L20:665 strongly suppresses
- L23:619 strongly excites
- L22:607 suppresses
- L18:440 excites
- L19:381 excites word-start 'acc' fragments
- L19:1853 excites
- L22:2898 excites
- L16:66 excites
- L0:1010 excites
- L0:2206 weakly excites
- L14:95 weakly suppresses
- L6:16 weakly excites
- L2:1661 weakly excites
- L21:55 weakly suppresses
- L16:73 weakly suppresses
- L0:716 weakly excites
- L1:2876 weakly suppresses
- L20:2984 weakly excites
- L17:2527 weakly excites
- L6:298 weakly suppresses
- L0:864 weakly excites
- L16:719 weakly excites
- L22:1155 weakly suppresses
- L16:524 weakly excites
Feeds later layers
None: this is the last layer.
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
- 横
- 箍
- eni
- 公益性
- GENCY
- xia
- 体
- 扫
- 楼层
- sag
Suppresses
- ent
- lor
- cccc
- lin
- ur
- lac
- lon
- over
- ombine
- lm