sentence-opening 'indeed' and similar
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. 16 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 connecting words that start a new point or add support, such as 'Indeed', 'Not only' and sometimes the end of a sentence just before one. The pattern is not very tight, and turning the neuron up did not produce a clear theme.
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.
-
they are not static. They can change profoundly throughout your life, for better or for worse, and one of the biggest influences on this change is your diet.⏎Indeed
-
, carbon dioxide, sulfur, and microorganisms. (Crude oil with lots of sulfur in it is called “sour,” as opposed to “sweet.”) If enough of these contaminants collect in a steel pipe, they’ll work together to
-
acy often decided to sacrifice the fate of other nations within the Republic.⏎This phenomenon occurred with varying intensity and, sadly, it is apparently being revived today. Not only in the form of Pope Francis’ unfortunate statements but also in the form of another
-
exercise, and plain-old-fashioned muscle fatigue. In addition, a lack of flexibility in the hamstrings is thought to spike the risk of hamstring injury (2). Indeed, research has suggested that athletes who have a history of hamstring injury have significantly less
-
tasks allowing your workforce to focus on more demanding tasks.⏎Also, these robots are very versatile and can quickly be relocated to another part of your production process. ndeed, compared to larger complex solutions such as traditional robots, cobots are quick to
-
beef requires 40 calories of fossil fuel energy, whilst producing one calorie of human-edible grain takes only 2.2 calories of fossil fuel energy4!⏎2) Preserve Habitats and Species⏎Eating animals 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: the fact that, in reality, the fact that the number of times that we can take the number of times we can take that number of times that number of times that number of times that number of times that number of times that number of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
added as we transcribe it.⏎- Front Matter⏎- A PREFACE containing a general Account of the Life, Character, and Writings of the AUTHOR: by John Clarke⏎- I. Of Moral Good and Evil⏎- II.
-
its disorders. In: Daroff RB, Fenichel GM, Jankovic J, Mazziotta JC, eds. Bradley's Neurology in Clinical Practice. 6th ed. Philadelphia, PA: Elsevier Saunders; 20
-
the the ptr array are interpreted as unsigned char for the comparison.⏎- Pointer to the block of memory where the search is performed.⏎- Value to be located. The value is passed
-
usion. In: Mason RJ, Murray JF, Broaddus VC, et al., eds. Murray and Nadel's Textbook of Respiratory Medicine. 5th ed.
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
- L14:1209 excites
- L9:12 suppresses
- L9:16 suppresses unclear: periods inside numbers and names
- L15:8 excites
- L13:212 weakly suppresses
- L14:602 weakly suppresses
- L10:144 weakly suppresses
- L12:731 weakly excites
- L11:112 weakly excites
- L15:1283 weakly suppresses
- L14:1187 weakly suppresses
- L4:96 weakly excites
- L14:451 weakly excites
- L5:1653 weakly excites
- L5:3 weakly suppresses
- L5:974 weakly excites
- L2:1321 weakly excites
- L6:1304 weakly suppresses
- L5:1371 weakly suppresses
- L12:653 weakly excites
- L13:9 weakly excites
- L8:1683 weakly suppresses
- L13:345 weakly excites
- L7:948 weakly suppresses
Feeds later layers
- L17:1394 weakly suppresses
- L22:3018 weakly excites
- L17:806 weakly excites
- L19:1638 weakly suppresses
- L18:959 weakly suppresses
- L20:2711 weakly suppresses
- L20:241 weakly suppresses
- L17:2852 weakly suppresses
- L22:2981 weakly suppresses
- L20:3021 weakly suppresses
- L19:2928 weakly suppresses
- L18:2242 weakly suppresses
- L23:2679 weakly excites
- L21:1594 weakly suppresses
- L21:578 weakly excites
- L25:1405 weakly excites
- L17:503 weakly suppresses
- L17:2285 weakly excites
- L19:2762 weakly suppresses
- L23:2362 weakly suppresses
- L21:724 weakly suppresses
- L18:2954 weakly suppresses
- L17:2499 weakly suppresses
- L17:1726 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
- (del
- rdf
- ński
- _PARAMETER
- rien
- enze
- 〘
- encing
- Raises
- plier
Suppresses
- correctly
- thăm
- 경우에는
- 括
- yyn
- uya
- 情况下
- 際に
- erva
- Brook