unclear: mid-phrase words in prose
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 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 responds to a range of ordinary words in the middle of sentences, such as 'used', 'teach', 'trust' and 'country', with no clear shared meaning. The strongest cases are mostly plain prose rather than one kind of text. Turning it up changed the wording of the model's replies but not in any obvious direction.
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.
-
Springs, Ark.⏎Uses of radon⏎Radon is sometimes used<|endoftext|>A Stepper Motor is a special form of AC Motor. The input of
-
are used in medicine to teach clinical skills, diagnostic reasoning and planning treatment of individual patients. The way of making
-
pires of Ghana, Mali, and Songhai, and slave exports had supplemented the export of gold.⏎Paraguay and Panama were the countries with the highest level of trust in the
-
electoral college in the United States, it assigns electoral-college votes to each country (including the United States), based on the size of its population. So,
-
the good work and remember any<|endoftext|>07 Nov Book Review: Jim Crow Terminals⏎Book Review: Jim Crow Terminals:⏎The Desegregation of American Airports⏎Jacob Anbinder, Harvard University⏎Anke Ortlepp.
-
in 'Ylam at the eastern bank of today's Shatt-il-Arab.⏎These remarks indicate two ingredients in the formation of early Mesopotamian⏎Christianity, that have merged gradually into the general Christian population.⏎But we can state
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 been thinking about the other one: the one: the one: the one: the one: the one: the one: the one: the: the one: the: the one: the one: the one: the:
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and western Asia, but it was introduced and naturalized to Australia, New Zealand, and North America over time.⏎Want to save this post for later? Click Here to Pin It On Pinterest!⏎Are you interested in learning more about this ancient medicinal
-
your next visit.<|endoftext|>Sometimes it makes sense for a function to return multiple things. In such cases, it's common to just return them in a tuple: "return (count, new_obj)".⏎Sometimes you might want to return objects of
-
This will disconnect your terminal from the server you are currently connected to.⏎DWAINE is the operating system of Research Mainframes. Scientists use it to run their research programs. Research Directors program guardbuddies on DWAINE.⏎DWA
-
.⏎That is a lot of hardware and it’s also a lot of history. NASA is worried that without some guidance and agreement irreplaceable relics of the Space Age, such as Neil Armstrong’s first footprint on the Moon could be lost and sites
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
- L14:2250 suppresses
- L14:2574 excites
- L14:1892 excites
- L14:110 excites
- L12:1373 suppresses
- L11:1588 excites
- L13:2917 excites
- L13:253 excites
- L13:1156 weakly suppresses
- L11:1122 weakly suppresses
- L13:397 weakly excites
- L11:2136 weakly excites
- L12:1712 weakly excites
- L13:219 weakly excites
- L14:1073 weakly suppresses
- L14:1638 weakly excites
- L14:2655 weakly excites
- L14:1434 weakly suppresses
- L13:1119 weakly excites
- L14:565 weakly excites
- L8:591 weakly excites
- L10:72 weakly suppresses
- L5:216 weakly excites
- L12:2946 weakly suppresses
Feeds later layers
- L16:3013 strongly suppresses
- L16:2393 strongly suppresses
- L16:1695 suppresses
- L18:2307 suppresses
- L16:1146 suppresses
- L18:761 suppresses
- L17:1323 suppresses
- L20:1595 suppresses
- L16:1578 suppresses
- L17:2271 suppresses
- L20:604 suppresses
- L21:349 suppresses
- L18:2205 suppresses
- L20:58 excites
- L19:2261 suppresses
- L16:2450 suppresses
- L18:1083 weakly suppresses
- L19:601 weakly suppresses
- L17:748 weakly suppresses
- L16:1474 weakly suppresses
- L20:484 weakly suppresses
- L17:614 weakly suppresses
- L17:1419 weakly suppresses
- L16:1108 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
- 把她
- 她说
- 告诉她
- 给她
- Gratuit
- ServiceImpl
- 主义者
- Giul
- 大发
- 零售商
Suppresses
- CD
- nd
- 0
- ometric
- cd
- los
- mp
- ert
- C
- ild