'see' in 'let's see' and 'to see if'
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 'see' when it is used as 'let's see' or 'to see if', meaning check or find out. Turning it up made the model drift into math-style text, which does not clearly match 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.
-
riting about those points to see which might make a useful topic. It allows them some room to explore potential topics
-
gave his recessive allele, they would have a 25% chance that their offspring would have this autosomal recessive trait or disorder.⏎So let's see what would happen if one parent was heterozygous and one parent was homo
-
looks at all the papers of all the students to see if having 89 identical answers is unusual, for students scoring the way John
-
through all of the answer options and eliminate choices that are obviously incorrect. In addition, it’s helpful to plug each answer option into the sentence and read it to see if it makes sense. Students who want to take advantage of these and other strategies
-
about it! Check out the brainteasers below to see the power of the brain in action, and answer the questions at the bottom of
-
Answer: Lines with the same slope are parallel and if the slope of one line is the negative reciprocal of the second line, then they are perpendicular.⏎Let's see how to identify parallel and perpendicular lines.⏎Let's take some examples to understand this
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: $ \frac{1}{n} $, $ \frac{1}{n} $, and $ \frac{1}{n} $, and $ \frac{1}{n} $, and $ \frac{1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of a risky asset is a random walk with drift, so its return is unpredictable."⏎4. Is the following statement true, false
-
answer specifies that the density function is⏎I got the f(x) = x-1 part, and I got how to calculate the variance after you have the expected values, but I'm lost on other questions.⏎My questions are:⏎Where do
-
tree-liana interactions: distribution and host preference of lianas in a tropical dry evergreen forest in India.⏎Vivek, P., & Parthasarathy, N.⏎Lianas are structurally dependent life-form that rely mostly on
-
implemented the law of giving a right to recall to the employers of certain industries. This law has been
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
- L3:2204 excites
- L3:810 excites
- L3:461 excites
- L2:1113 weakly excites
- L3:1793 weakly suppresses
- L1:1910 weakly excites
- L0:1346 weakly excites
- L1:2858 weakly suppresses 'as' in Incas, 'te' in intelligence
- L3:3064 weakly excites
- L0:903 weakly excites
- L3:2673 weakly suppresses
- L2:1036 weakly excites
- L2:2653 weakly excites
- L3:388 weakly excites
- L1:29 weakly suppresses
- L1:782 weakly suppresses
- L2:75 weakly suppresses
- L3:777 weakly suppresses
- L3:1082 weakly suppresses
- L3:1119 weakly suppresses
- L0:170 weakly excites
- L1:898 weakly suppresses
- L2:126 weakly excites
- L3:2722 weakly suppresses
Feeds later layers
- L5:1690 suppresses
- L19:1511 suppresses
- L19:1753 weakly suppresses
- L20:826 weakly suppresses
- L23:1428 weakly suppresses
- L23:1981 weakly suppresses
- L23:2567 weakly suppresses
- L6:2031 weakly suppresses
- L5:740 weakly suppresses
- L12:682 weakly excites
- L13:2122 weakly suppresses
- L7:1006 weakly suppresses
- L5:2212 weakly suppresses
- L6:697 weakly suppresses
- L8:77 weakly suppresses
- L7:2720 weakly suppresses
- L7:1072 weakly suppresses
- L22:582 weakly suppresses
- L23:588 weakly suppresses
- L18:2836 weakly excites
- L15:2067 weakly excites
- L7:2277 weakly suppresses
- L9:1622 weakly excites
- L13:2469 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
- affe
- ARN
- 胱
- rame
- ramework
- 学校
- ARSE
- Phar
- alte
- search
Suppresses
- izer
- 跌破
- izing
- .databind
- :";⏎
- decreasing
- 她说
- Specifications
- creasing
- -legged