unclear: words that set up a qualifier
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. 13 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 fires on small words that set up a qualification or a contrast, like 'may', 'but', 'conversely' or 'in the carefully selected patient'. Turning it up made the text wander into unrelated topics, so it does not clearly control anything visible.
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.
-
precisely remove material to produce two smooth surfaces that are parallel with each other and to the hub-mounting surface. The problem with thickness variation is that too much material may need to be removed to establish these surfaces, leaving the rotors below the minimum
-
of sunscreen have minimized the amount of vitamin D that many of us get from sun exposure. Calcium is essential for strong bones and a healthy heart, but too much is not good.⏎In fact, excess calcium, which the NIH describes as more than
-
eight weeks, which isn't always "just gas." By then, your baby will show pleasure at the sight and sound of their parents, and may get sad when they don't see them.⏎- Step 6: Look for cause and effect
-
yet most biological movements involve a large number of degrees of freedom (in a rigid arm this number is directly related to the number and type of joints). An extreme case with especially high degrees of freedom is given by systems in which muscles are unattached
-
30. It tells you the oil’s thickness or viscosity. A thinner motor oil — indicated by a lower number — flows more easily. Conversely, a thicker oil — with a larger number — is thick and resistant to flow.
-
method in which the muscles and tendons are stretched around the fractured bone to help align it. Surgery for a broken bone isn’t always required, but may be needed to support and position bone fragments.⏎Hernando Orthopaedic provides expert fracture
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 working with the code, to let me to the to the... I... I... to the... for to... in a way to... to the... to... in a way to... to the... to... to...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
caves at the Wonderland of Rocks, take the place seriously. The average altitude is 4,400 feet and there are dramatic temperature extremes between day and night. Bring layers to dress up and down for temperature shifts. The terrain is more
-
should also considered. This is the trick you should apply. Do you want a garden contoured hill (hill), normal (flat), terraced, or variations? But remember, everything should fit your needs.⏎Afterwards, do not forget to
-
ford breed which was founded in the mid-18th century by the farmers of Hereford County, England. Among the horned Herefords an occasional calf would be born which did not develop horns. This change from parents' characteristics is
-
point about your example or thing. For example, if you are writing up a newspaper about mobile phones, you might be requested to compare sales reports from two different models. Similarly, if you’re analyzing the affects of smoking on children’s asthma,
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
- L12:2041 suppresses
- L12:2281 excites
- L12:764 suppresses
- L11:1960 suppresses
- L11:841 suppresses
- L11:998 excites
- L12:1356 suppresses
- L12:836 suppresses
- L12:2016 suppresses
- L12:690 excites
- L12:2917 excites
- L12:1991 suppresses
- L11:999 excites
- L12:1894 excites
- L12:1280 suppresses
- L12:122 excites unclear: word before a clause continues
- L12:798 excites
- L12:2942 suppresses
- L12:105 excites
- L12:683 excites
- L11:644 suppresses
- L12:1248 excites
- L8:135 excites
- L12:913 suppresses
Feeds later layers
- L14:2536 suppresses
- L14:2206 excites
- L14:1740 suppresses
- L15:772 excites
- L15:1446 excites
- L14:2655 excites
- L15:1790 suppresses
- L14:1686 suppresses
- L16:3059 excites
- L15:2010 excites
- L14:346 excites
- L14:1161 excites
- L14:1756 excites
- L14:2223 excites
- L14:2771 excites
- L14:2875 suppresses
- L14:2307 excites
- L14:2873 suppresses
- L15:2399 suppresses
- L14:1795 excites
- L14:482 excites
- L14:1150 excites
- L16:2265 excites
- L15:1705 excites
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
- otation
- otope
- ination
- inations
- 疹
- oin
- inesis
- ][]
- okers
- ranges
Suppresses
- aq
- UCE
- >",
- >Action
- aklı
- 优越
- ba
- !",
- 蔼
- iệu