"so" opening a purpose 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. 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 fires on the word "so" when it introduces a purpose, as in "so that" or "so as to". Turning it up made the model repeat "that as as as", and the negative side is weak and mostly on "make" in phrases like "built to make".
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.
-
adequacy of reservation for them. The major hurdle for the effective dispensation of social justice is the lack of data on sectoral population, and their due representation so far.⏎The identity of caste degradation is permeated in the socio, economic and
-
treats of the best means of directing school affairs so as to secure efficiency. It includes not only school economy, but also school discipline
-
diseases to a certain level.⏎Microbial science is the detailed study of microorganisms that are tiny and it is important to understand the performance of every microorganism so that better understanding can be generated towards the living bodies. Microbial sciences help in
-
the building. For example, if your building is minimalist concept, then create a park with the same concept. Garden design should fit with the concept of the building so that it becomes a harmonious unity⏎Well, the next step is clean the
-
These are the commands, decrees and laws the Lord your God directed me to teach you to observe in the land that you are crossing the Jordan to possess, so that you, your children and their children after them may fear the Lord your God
-
two pieces, it is possible to fit the pieces together again to make a new shape. How many new shapes can you make?⏎Can you shunt the trucks so that the Cattle truck and the Sheep⏎truck change places and the Engine is
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 am going in the morning, and the house is still still off the house. It is a very quiet house of the moon, and the house of the planet. The planet of the earth is of the house of the planet. The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and showered us with love last year.⏎Today, we are posting the Economics– Important Questions Bank for Rajasthan Board HSC 2016 Examination to make life easy for all you HSC students.⏎Without making you wait any further,
-
price is one of the most critical procedures for bond investors, as miscalculating can lead to huge losses. But worry not. This calculator is built to make the process easier for you.⏎We have written this article to help you understand what
-
alveolus. The alveoli are tiny air sacs within the lungs where the exchange of oxygen and carbon dioxide takes place.⏎Last Editorial Review: 6/14/2012⏎Back to MedTerms online medical
-
before the party, we do not have known or seen the previous Congress, the Indian National Congress in the fight for India’s freedom. INC was established in the year 1885 the Theosophical Society of Indian and British members.
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
- L19:1228 excites
- L10:190 excites
- L18:477 excites
- L18:177 suppresses
- L13:1056 suppresses end of document boundary tokens
- L9:812 excites
- L9:429 excites
- L10:556 weakly excites
- L19:119 weakly suppresses
- L18:1018 weakly suppresses
- L15:59 weakly excites
- L10:1289 weakly suppresses
- L0:323 weakly suppresses
- L1:2401 weakly suppresses
- L9:146 weakly suppresses
- L19:2535 weakly excites
- L1:558 weakly suppresses
- L1:68 weakly excites
- L10:1032 weakly excites
- L6:1354 weakly excites
- L17:257 weakly excites
- L19:3040 weakly excites
- L0:480 weakly suppresses
- L3:59 weakly excites
Feeds later layers
- L21:1111 excites
- L21:2876 excites
- L23:914 excites
- L22:1186 excites
- L26:586 excites
- L27:616 weakly excites
- L24:306 weakly suppresses
- L22:631 weakly excites
- L23:1754 weakly suppresses
- L27:291 weakly excites
- L21:1541 weakly suppresses
- L24:36 weakly excites
- L21:1495 weakly excites
- L25:1722 weakly excites
- L22:2960 weakly suppresses
- L24:296 weakly suppresses
- L25:246 weakly suppresses
- L21:1352 weakly suppresses
- L22:1798 weakly suppresses
- L23:1860 weakly suppresses
- L23:2575 weakly suppresses
- L24:2157 weakly excites
- L24:627 weakly excites
- L24:840 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
- rằng
- that
- _that
- →that
- että
- that
- asel
- ersed
- 、
- alion
Suppresses
- 瓒
- pNet
- atica
- очно
- OfClass
- -off
- очных
- Druid
- moon
- went