"that" after "such" or "such a way"
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. 4 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 "that" when it follows "such" or "in such a way," as in "numbers such that" or "so that." It is mostly used in maths and logic sentences. Turning it up did not produce a clear theme in the model's replies.
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.
-
) where they proved that a(n) is not more than the smallest number k such that the total number of ones in the binary expansion of numbers from 1 to k
-
Where is the liturgy celebrated?<|endoftext|>A particle is moving along the parabola 4y = (x + 2)^2 in such a way that its x-coordinate is increasing at a constant rate of 2 units per second.
-
77, the sequence of numbers such that their powers of 2 do not contain zeros: 1, 2,
-
holeLanguage brain teasers are those that involve the English language. You need to think about and manipulate words and letters.⏎Enter words into the following word chain such that each pair of words in the chain forms a compound word. No word can appear
-
and the operation of convolution define a commutative ring. Any ring without divisors of zero can be extended to a quotient field or field of fraction such that⏎The convolution quotient of a function with itself; i.e., f/f;
-
a Liki-design? To get an idea try to copy one of the Liki-designs on squared paper in such a way that each unit square has only one colour. What can be said about the distribution of
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: a very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very, very
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
real art” only comes from the West. Introductory discussions of the history of art history and geographic conventions set the stage for considering the larger question, “where does art come from,” and other questions such as, “why haven’t I heard
-
. carbon emissions in half by 2030 and doing so equitably.⏎But neither plan explicitly grapples with the politically thorny question of where to invest. Should the money follow the people, who have flocked for decades to
-
that those drawbacks could be cured by sound engineering.⏎And so he set himself the task of creating and manufacturing a revolutionary new automobile, the Hughes Steamer.⏎Where to start?⏎Howard
-
made news when it was shut down for building code violations. Part of the center’s role was to provide accommodations for clients to “do the month.” So where do these postpartum rituals come from and are they good for new moms?⏎Researchers
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
- L3:1390 excites 'that' in 'so that' purpose clauses
- L3:1953 excites 'such' after 'no' or 'any'
- L3:2501 weakly suppresses
- L1:2738 weakly suppresses
- L3:2598 weakly excites
- L3:866 weakly excites
- L3:2069 weakly suppresses
- L3:30 weakly excites
- L3:2638 weakly excites
- L1:1029 weakly excites
- L2:1603 weakly excites
- L3:572 weakly excites 'that' after 'so ... that'
- L3:1763 weakly excites
- L1:1024 weakly excites
- L3:239 weakly excites
- L1:1475 weakly suppresses
- L3:1277 weakly suppresses
- L1:2629 weakly suppresses
- L2:2692 weakly excites
- L1:1069 weakly suppresses
- L2:2929 weakly suppresses
- L3:2952 weakly excites
- L1:1830 weakly suppresses
- L3:2037 weakly excites
Feeds later layers
- L6:2466 excites
- L6:1199 excites
- L7:1673 excites
- L18:2056 excites
- L19:2437 excites
- L9:932 weakly excites
- L6:830 weakly excites
- L17:2225 weakly suppresses unclear: verbs after subject, line breaks
- L11:1470 weakly suppresses
- L14:1057 weakly excites
- L8:2635 weakly excites
- L21:1879 weakly excites
- L13:895 weakly excites
- L5:1064 weakly excites
- L18:1612 weakly excites
- L8:132 weakly excites
- L5:2727 weakly excites
- L15:2416 weakly excites
- L6:1304 weakly excites
- L14:2405 weakly suppresses
- L6:1663 weakly excites
- L15:504 weakly excites
- L7:186 weakly suppresses
- L17:2671 weakly excites the word matter in no matter phrases
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
- positivity
- positive
- interesse
- positively
- ParameterValue
- Positive
- 有条件的
- Positive
- passed
- posit
Suppresses
- atures
- 씩
- 也不错
- 舐
- ths
- '),'
- mits
- تين
- 也将
- 也知道