comma closing a main clause
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. 1 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 responds to commas that come right after a complete statement, where the sentence is about to add an extra detail. It is a weak signal. When turned up, the model's replies wandered into unrelated topics, so the steering result does not confirm the pattern.
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.
-
precarious world. During attacks, the rural populace could flee behind city walls, where defence forces assembled to repel the enemy. The wall served this purpose for millennia, until the invention of heavy artillery rendered walls useless in warfare. With the advent of
-
Berkeley. It’s a comprehensive resource for learning more about the processes of science as used in the real world. The processes of science are represented as fluid and iterative, showing that science does not follow a structured recipe. The three
-
istic period. Known for their strongly realistic portraits, this tetradrachm of Eukratides is no exception. He is depicted as a formidable warrior, wearing a cuirass and a bronze helmet over his diadem. His features
-
how hard the local population work, they remain the poorest of the poor.⏎The Tam Nông Hospital is an old hospital built in the early 80s, originally intended for 20 beds.⏎Tam Nông Hospital⏎Due to the
-
multi-core paradigm. The microprocessor has proven to be a versatile invention, branching out into numerous fields outside the personal computer.⏎Microprocessors have revolutionized the world, especially in the area of electronics. A myriad of modern items ranging from cell phones
-
and Lorraine families. The Medici collection began in the 16th century, first in Palazzo Vecchio, then at the Uffizi and finally in
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 have a 4-digit number, the number of 3-digit numbers, and the number of 2-digit numbers. Maybe, I have a 1-digit number. Perhaps, I have a 0-digit number. So, I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
implemented, for example, by the laws of King Benjamin and Mosiah. Yet they were constantly reminded that salvation does not come by the law of Moses:⏎“And, notwithstanding we believe in Christ, we keep the law of Moses, and look forward
-
roads are not necessarily the best solution.The statistics on car ownership and per capita income suggest that planners should start to think in a different way about road transport – and, indeed, about air transport – in Papua New Guinea. For the foreseeable future,
-
called impersonal verbs.⏎An impersonal verb is one that has no determinate subject.⏎Sometimes the subject is not represented at all; as, Methinks, meaning, it seems to me; lists, meaning, it ple
-
the class and helping the class become a successful place of learning.⏎The instructor presented a survey to the students on the first day about the merit of participation; and, how participation can affect the class.⏎The first function of the research plan focuses on
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
- L0:1301 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L0:1596 weakly excites
- L0:313 weakly suppresses "and" joining paired nouns
- L0:68 weakly suppresses
- L0:925 weakly excites
- L0:1961 weakly suppresses
- L0:80 weakly excites
- L0:2195 weakly excites
- L0:1595 weakly excites
- L0:1001 weakly suppresses
- L0:2479 weakly suppresses
- L0:573 weakly suppresses
- L0:2296 weakly excites
- L0:2573 weakly excites
- L0:1260 weakly excites
- L0:148 weakly suppresses
- L0:98 weakly suppresses
- L0:1753 weakly excites
- L0:1497 weakly suppresses
- L0:1000 weakly excites
- L0:2475 weakly suppresses
- L0:2005 weakly excites
Feeds later layers
- L16:695 weakly excites
- L14:1639 weakly excites "and" joining two nouns
- L11:2039 weakly excites
- L13:0 weakly excites
- L16:450 weakly excites
- L16:1489 weakly excites
- L6:1937 weakly suppresses
- L16:42 weakly excites
- L18:2956 weakly suppresses
- L2:12 weakly suppresses
- L13:491 weakly excites
- L16:319 weakly excites
- L6:2410 weakly suppresses
- L10:2268 weakly suppresses
- L20:2731 weakly suppresses
- L18:12 weakly excites
- L18:54 weakly excites
- L10:95 weakly excites
- L16:480 weakly excites
- L13:518 weakly suppresses
- L9:661 weakly excites
- L19:1541 weakly excites
- L4:2903 weakly excites
- L2:2367 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
- dere
- others
- jq
- kür
- rés
- _PHP
- izable
- 大国
- doubleValue
- .TextAlign
Suppresses
- emos
- elines
- ��
- `,
- },{
- ination
- abol
- nữa
- 께
- .),