"and" or "or" in a pair or list
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. 15 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 word "and" (or "or") when it joins two or three things of the same kind, like "silver, tin, and copper" or "men and women". Turning it up made the model repeat phrases, and turning it down gave a more general, flowing style.
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.
-
two ends on the same continuum?⏎Homer, Plotinus, and Plato have all posited answers to this question, whether directly - in the case of Plato and Plotinius - or through story - in the case of Homer.⏎Plotinus
-
"silver" are actually a mixture of silver, tin, copper, and mercury, and are more properly called amalgam fillings. The silver, tin, and copper are provided in a powder form by various manufacturers, with silver being the major
-
. In the late nineteenth century a Scottish Mill designed and produced four cloths, one for each of the world’s most distinguished universities. While Harvard, Yale, and Cambridge all had their respective cloths lost to history, for reasons unknown Oxford endured
-
the FCAT?⏎The FCAT includes multiple-choice, gridded-response (fill in the blanks) and performance tasks (such as essays). The multiple-choice and gridded-response questions are machine scored. Each performance task test is scored by
-
up.⏎- Wear clothing that will allow the skin to breathe, so you don’t become overheated.⏎- A hat helps to hold the heat in during winter and protects the eyes in the summer.⏎The way you hold the body while running is
-
or more for women. The study found that of the almost 50,000 people surveyed across China, 55.6% of men and 15% of women were current drinkers, having had at least one drink
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: how to use the single, single, and the single, in the single, single, and in, on the single, without, on the single, the single, and all, in the single, and, by, on the single
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, movies and other licensed digital information resulting in loss of potential revenue. Examples include, attack on the national infrastructure such as electricity grid, electricity and water, delivery system, air traffic controls that depend on communications and network. This
-
negative. You blood groups help doctors know many things, like what kind of blood you’ll need in a blood transfusion or an organ donation, and if you or your baby will be in danger if you get pregnant.⏎So how does your body
-
in a discontinuity analysis on observations near the borders between anglophone and francophone countries.⏎- Colonial rule⏎ASJC Scopus subject areas⏎- Economics and Econometrics⏎- Organizational Behavior and Human Resource Management<|endoftext|>- About ASTI
-
105° Fahrenheit).⏎How can I prepare?⏎- Listen to local weather forecasts and stay aware of upcoming temperature changes.⏎- Be aware of both the temperature and the heat index. The heat index is the temperature the body feels when the effects
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
- L6:427 suppresses
- L12:2058 excites
- L14:2202 excites
- L14:801 excites 'and' or 'or' joining list items
- L12:1786 excites
- L7:1223 suppresses "and" joining two decades or times
- L12:51 excites
- L14:2865 suppresses
- L9:92 suppresses "and" or "or" joining two items
- L2:1000 excites
- L13:1554 excites
- L11:406 weakly suppresses
- L12:1336 weakly suppresses
- L10:2046 weakly suppresses
- L9:738 weakly suppresses
- L1:1250 weakly suppresses
- L9:2010 weakly suppresses
- L9:77 weakly excites commas in citations and name lists
- L4:57 weakly suppresses
- L13:2931 weakly excites
- L13:131 weakly excites 'and' joining parallel phrases
- L5:880 weakly excites
- L7:836 weakly excites
- L12:2322 weakly suppresses citation and catalogue metadata text
Feeds later layers
- L16:1937 excites 'and' joining two time or quantity ranges
- L16:2454 weakly suppresses
- L18:2354 weakly excites
- L16:2577 weakly excites
- L16:2940 weakly excites commas inside lists and names
- L16:1805 weakly excites
- L16:2145 weakly excites
- L16:2568 weakly suppresses
- L16:1692 weakly suppresses
- L17:2623 weakly excites
- L16:279 weakly suppresses
- L20:364 weakly excites
- L18:1975 weakly excites
- L18:2385 weakly suppresses
- L17:1610 weakly suppresses
- L20:1009 weakly excites
- L23:1275 weakly suppresses
- L20:1458 weakly suppresses
- L18:2037 weakly suppresses
- L25:2927 weakly excites
- L16:2296 weakly excites
- L25:1758 weakly excites
- L20:901 weakly excites
- L17:1022 weakly 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
- idad
- itate
- itating
- We
- osci
- we
- ista
- razy
- itated
- otle
Suppresses
- 骝
- 的支持
- FE
- 告诉她
- Registration
- igner
- 另一边
- aar
- 叮
- Qualität