unclear: word before a company or brand name
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. 19 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 seems to fire on the word just before a company, product or brand name, such as 'Space' in a company name or 'code' before a promo code. The pattern is loose, and turning it up gave no clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
when he saw Halley's comet in 1986. Now he travels around Namibia, providing stargazing and astronomy education through his company, Space Observation Learning Namibia. His goal is to have more young heads in the clouds
-
earlier convoy battles.⏎An effort to improve the Luftwaffe support<|endoftext|>Extra 10% off sitewide + Free shipping over $150 with code: STOCKUP Details⏎Happy Homeschool Month! The whole month of January we
-
you may want to know more about the lifespan of the trust. The Houston living trust lawyers at Mendel Law Firm, L.P. explain when and how a living trust terminates
-
. Several later models of the Omnichord added MIDI compatibility, a greater selection of sounds for the Sonic Strings, vibrato, and chord memory, called Chord Computer.⏎The OM27 was the first Omnichord model, released in
-
-class dentistry today.⏎If you have questions or concerns about dental technology or just want to come take a look at all our cool machines, get in touch with Creative Dental Solutions today!⏎The content of
-
in Montreal have won $1 million in seed money for developing a protein-rich flour made from grasshoppers. The students call their invention 'Power Flour', and their achievement underscores the nutritional challenges posed by an expanding world population.
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: the The The The The The The The The The The The The The The The The The The The The The The The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
900 – 14 August 1958) was a French physicist, who shared the 19
-
states it as being restless. http://volcansvanuatueruptionsgb.blogspot.fr/2014/03/08032014-nevado-de-colima-cleveland.html and used this image
-
-4369⏎DOMINGUES, Beatriz Helena. The guarani jesuitic missions in the context of Enlightenment. História [online]. 2006, vol.25, n.1, pp.
-
ivative Works 4.0 License.⏎Spreadsheets: Laying a Foundation for Understanding Functions,⏎Spreadsheets in Education (eJSiE):⏎2, Article 1.⏎Available at: http://epublications.bond.edu.au
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
- L18:151 strongly excites
- L16:1129 strongly suppresses
- L18:2453 strongly excites
- L15:1219 suppresses
- L18:781 suppresses
- L16:498 suppresses
- L15:2839 suppresses
- L15:1272 excites
- L17:1840 suppresses
- L16:248 suppresses
- L15:434 suppresses
- L13:1777 suppresses
- L15:1317 suppresses
- L18:1077 suppresses
- L16:2381 excites
- L10:812 suppresses
- L15:1607 excites
- L16:370 suppresses
- L16:1262 suppresses
- L13:450 excites
- L18:129 excites
- L14:970 suppresses
- L13:2946 excites
- L5:2917 suppresses
Feeds later layers
- L20:52 strongly excites
- L20:33 excites
- L20:573 suppresses
- L21:105 suppresses
- L20:312 suppresses
- L20:19 excites
- L21:12 excites
- L25:5 excites
- L20:669 suppresses
- L22:7 excites
- L20:323 suppresses
- L20:38 excites
- L21:142 excites
- L20:2065 suppresses
- L20:151 suppresses
- L22:151 suppresses
- L23:3 excites
- L23:206 suppresses
- L25:3 excites
- L20:17 excites
- L21:475 suppresses
- L21:24 excites
- L22:29 excites
- L25:409 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
- __
- <|endoftext|>
- high
- H
- [
- 固
- [
- H
- low
- leston
Suppresses
- UD
- ?',
- Porn
- ベル
- Bí
- кт
- bang
- 起来了
- física
- .shtml