unclear: fractions and odd characters mid-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. 21 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 reacts to unusual symbols: fraction signs like ⅕ and ⅔, circled numbers, runes, and places inside foreign-script words where a letter or accent seems to be missing. The pattern is hard to summarise in one idea, and steering did not produce a 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.
-
weeverfish, adder pike, black fin, little weever.⏎It has an elongated body, compressed at the sides; its height exceeds ⅕ of the body length. It is covered with small cycloid scales; the
-
The angular velocity of the gear blank is ω 1 and ω 1, and the angular velocity of the gear blank is ω 1;⏎② The blade moves with the revolution of (q) plane, and the angular
-
do it!⏎- Bottle (an empty plastic water bottle will work)⏎- Vegetable oil⏎- Food coloring⏎- Alka-Seltzer tablets⏎Fill ⅔ of the bottle with vegetable oil, and fill the remaining part with water.
-
ic inscription in Younger Futhark:ᛒᚱᚢᛋᛏᚢᛚᛁᚴᛅ⏎Younger Futhark runes were used from
-
rx’r [Lurje, 2004, p. 20]. In the New Persian language this word is transferred in form ﺭﺎﺨﺭﻓ farxār [Baev
-
skiy, 1980, p. 88] or ﺭﺎﻬﺑ bihār [Hudūd al-‘
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 I was the result of a **moral and natural** **centuries**? What is the result of me—? What is the result of me? What would I be? What would I be—? What would I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
hardship.⏎This I can say for myself:⏎that for awhile I was the Heodeninga<|endoftext|>This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.⏎As the soil, however
-
for<|endoftext|>- Agriculture in NSW is important for the state’s economy, industries, communities and
-
and the authors also include methods that allow for the solving of nonmathematical problems as mathematical problems.⏎To view this DRM protected ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software.
-
, toward positive holistic wellbeing.⏎Strengthen, support and guide stakeholders, to further advance the self-determined education aspirations, achievements and successes for Kahungunu ākonga, whānau, hapū and iwi.⏎Design
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
- L19:2973 suppresses
- L18:528 suppresses
- L17:2096 suppresses
- L18:1940 excites
- L17:1333 suppresses
- L18:480 suppresses
- L19:377 suppresses
- L19:1280 suppresses end-of-document token between unrelated texts
- L19:771 suppresses
- L6:1374 weakly suppresses
- L4:32 weakly excites
- L16:1046 weakly excites
- L15:643 weakly excites
- L18:901 weakly suppresses
- L19:1167 weakly excites
- L16:442 weakly suppresses
- L16:1637 weakly suppresses
- L9:808 weakly suppresses
- L3:1324 weakly excites
- L3:298 weakly suppresses
- L3:923 weakly suppresses
- L16:1438 weakly suppresses
- L10:1630 weakly suppresses
- L7:1321 weakly excites
Feeds later layers
- L23:2636 excites
- L23:2273 excites
- L26:1792 excites
- L23:166 weakly suppresses
- L26:893 weakly excites
- L22:320 weakly excites
- L23:2050 weakly suppresses
- L26:747 weakly excites
- L25:429 weakly excites
- L22:2310 weakly suppresses
- L27:674 weakly excites
- L23:2151 weakly suppresses
- L22:1328 weakly excites
- L23:364 weakly suppresses
- L27:2688 weakly excites
- L24:1568 weakly suppresses
- L22:1854 weakly suppresses
- L22:2049 weakly suppresses
- L22:1485 weakly suppresses
- L23:279 weakly suppresses
- L23:2283 weakly suppresses
- L25:3012 weakly excites
- L22:831 weakly suppresses
- L27:1670 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
- →→→→→→→→→→→→→→→→
- →→→→→→→→→→→→→→→
- →→→→→→→→→→→→→→→→→→→→
- →→→→→→→→→→→→→→→→→→
- →→→→→→→→→→→→→
- →→→→→→→→→→→→→→→→→
- →→→→→→→→→→→→→→
Suppresses
- elight
- dera
- liền
- гон
- lier
- prem
- Sala
- iss
- ons
- próp