"in" in "which in turn" phrases
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. 20 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 lights up on the word "in" in phrases like "which in turn" or "and in return", where one thing leads to another. It goes quiet, and slightly negative, on "on" or "the" in phrases like "on the other hand" or "on the part of". Turning it down made the model repeat "on the part of" and "on the other hand" over and over, while turning it up gave loose, unrelated text.
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.
-
. The trees that die as a result of rising water levels attract insects, which in turn feed woodpeckers, whose holes later provide homes for other wildlife.⏎
-
an interaction of nature, nurture and the decisions and choices that people make. Biological programming is your genetics. Genes provide the instructions for structuring amino acids which in turn influence the proteins within our body cells thus programming chemical basis of our biology.
-
the group changed its name to the STEMG.⏎The STEMG reports to the ASD Product Support Specification Group (PSSG) which in turn reports to the ASD Services Commission (SVC).<|endoftext|>Biology exquis
-
production will be the most affected by climate change by 2030. This would lead to food insecurity, which in turn could lead to social instability, as was already the case in Mozambique in
-
is an example of mutualism because the nectar from the flower nourishes the bee and in return, the bee pollinates the flower.⏎Activity Three: Humans and the
-
the initial state of the atmosphere is critical. Even if Doppler radar information may seem highly localized, we find that it enhances the regional atmospheric conditions, which, in turn, can significantly improve the dynamic prediction of how the monsoon depression will move
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: on the one hand, on the part of the part of the part on the part on the part of the part part on the part on the part part on the part on the part on the part on the part part on the part on
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
his responsibilities, and the means by which his duties are to be performed. One of the chief causes of failure in teaching is the lack of well-defined purpose on the part of those entrusted with the training of pupils. A knowledge of the end to
-
. Hard work can lead to a deep sense of accomplishment and mastery, but can also be time-consuming and may not be the most effective approach. Smart work, on the other hand, can save time and increase productivity, but may lead to a
-
to perform specific functions.⏎Microcontrollers are designed to handle simple tasks, making them a popular choice in low-cost, low-power embedded systems. FPGAs, on the other hand, are designed to handle complex logic functions that require high performance and
-
to achieve and have clear steps and timelines for completion. They are usually long-term and require dedication and hard work in order to be achieved. Resolutions, on the other hand, are more short-term commitments that a person makes in order to improve
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
- L19:2530 strongly excites
- L19:934 strongly excites
- L18:2007 excites
- L15:2777 weakly suppresses
- L7:2179 weakly excites
- L19:1644 weakly suppresses
- L3:1289 weakly suppresses
- L7:3031 weakly suppresses
- L6:17 weakly suppresses
- L19:417 weakly excites
- L16:2008 weakly excites
- L11:43 weakly suppresses
- L18:569 weakly excites
- L17:1259 weakly suppresses
- L19:212 weakly excites
- L10:17 weakly suppresses
- L9:8 weakly excites
- L19:2308 weakly excites
- L15:2926 weakly excites
- L17:2541 weakly suppresses comma before but also in not only constructions
- L17:2988 weakly excites
- L7:934 weakly suppresses
- L11:1253 weakly suppresses
- L11:72 weakly excites
Feeds later layers
- L27:177 strongly excites
- L27:721 strongly suppresses
- L26:951 strongly suppresses
- L24:1796 strongly suppresses
- L23:1438 suppresses
- L26:2330 excites
- L25:2282 suppresses
- L27:1194 suppresses
- L25:2569 suppresses
- L24:1898 excites
- L23:2742 suppresses
- L25:666 suppresses
- L21:1237 suppresses
- L24:1785 weakly suppresses
- L25:1892 weakly suppresses
- L24:1669 weakly excites
- L21:2136 weakly excites
- L23:476 weakly suppresses
- L24:1141 weakly suppresses
- L27:2013 weakly suppresses
- L23:2234 weakly suppresses
- L22:1190 weakly excites
- L22:1368 weakly excites
- L22:2554 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
- 逵
- ��
- daf
- lotte
- EPROM
- 茴
- congratulations
- 櫃
- ottes
- 熟悉的
Suppresses
- on
- On
- On
- .on
- →on
- _on
- →On
- on
- -on
- (on