plural '(s)' in parentheses
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 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 bracketed '(s)' that writers add to words like 'agent(s)' or 'pin(s)' when they are not sure whether something is singular or plural. When it was turned up, the model's writing drifted into odd, repetitive text that had nothing to do with the topic.
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.
-
nine were found to be “obstructive.”<|endoftext|>Rural earthen roads impact assessment in Phewa watershed, Western region, Nepal⏎© The Author(s). 2016⏎Received: 23 November 20
-
to stimulation or verbal commands. It includes administration of sedative and/or analgesic agent(s) by a route other than IV; (PO, PR, Intranasal
-
17.0.0/17⏎- We need a network with <|endoftext|>|Classification||Electronic musical instrument, synthesizer|⏎|Inventor(s)||Suzuki Musical Instrument Corporation|⏎|Tronichord, autoh
-
’s airway, protective reflexes and the ability to respond to stimulation or verbal commands. It includes intravenous administration of sedative and/or analgesic agent(s) and appropriate monitoring.⏎Local Anesthesia: The loss of pain sensation over a
-
to 3.66 m) in width, and 8 feet to 12 feet (2.44 m to 3.66 m) in height. These measurements are common wall and roof dimensions in building design.
-
. The drops must be warmed to body temperature by putting the bottle in your hand for a few seconds before use. Before administering the drops, washing the affected ear(s) and drying them with a clean towel is necessary. To administer
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 problem of the same problem, and the problem of the same problem, and the problem of the same problem, and the problem of the same problem, the same problem, all the problem of the same problem, all the problem of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ter, Taslima and Simonovic, Slobodan P., "A General Overview of Multi-objective Multiple-participant Decision Making for Flood Management" (2002). Water Resources
-
the Australian population (Australian Journal of General Practice (2021) but many of us don’t really understand what this diagnosis
-
1. Which means they proved that a(30) ≤ 16.Share:<|endoftext|>Communication for Empowerment⏎Communication for Empowerment (C4E) is one of the practical approaches developed by OGC to address
-
groundwater for hydro electric power generation⏎Bardsley, W.E. (1995). Using groundwater for hydro electric power generation. Journal of Hydrology (NZ), 34(1), 1-14.⏎Permanent Research
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
- L10:107 strongly suppresses unclear: spots where text was removed
- L7:701 excites
- L15:1374 excites
- L11:1579 suppresses
- L3:1388 excites
- L10:142 suppresses opening parenthesis mid-sentence
- L15:650 excites
- L1:102 excites
- L9:632 suppresses
- L4:1572 excites
- L3:10 excites
- L10:43 excites "terms" in "in terms of"
- L5:2525 suppresses
- L12:50 suppresses
- L15:2795 suppresses
- L15:608 suppresses
- L9:982 suppresses
- L6:293 suppresses
- L13:2197 suppresses
- L6:468 excites
- L8:62 suppresses
- L8:79 suppresses
- L9:2249 suppresses
- L11:2783 excites
Feeds later layers
- L19:471 strongly excites
- L18:678 strongly excites
- L20:131 excites
- L20:306 excites
- L19:353 excites
- L17:776 excites
- L22:765 excites
- L17:70 excites
- L23:191 excites
- L20:597 excites
- L22:161 excites
- L24:508 excites
- L20:892 suppresses
- L21:189 excites document break token mid-sentence
- L25:17 excites
- L19:278 excites
- L26:326 excites
- L19:204 excites
- L24:39 excites
- L24:392 excites
- L26:836 weakly excites
- L23:517 weakly excites
- L23:290 weakly excites
- L18:2164 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
- oplevel
- condolences
- Equality
- ichten
- anken
- relationship
- ้ว
- 값을
- ProcessEvent
- vows
Suppresses
- Cop
- !)
- !)⏎⏎
- ège
- Sophie
- Ģ
- 预
- yat
- би
- Mart