unclear: mixed word fragments
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. 25 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
Nothing clear ties this neuron together. It fires on a grab bag of word pieces in very different topics, such as a Dutch-Afrikaans name, a geology term, an engineering term, and a welding term. Turning it up made the model ramble in broken, repetitive text.
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.
-
Voortrekkers comprised two groups from the eastern frontier region of the Cape Colony, semi
-
is obtained from drainage observations and a surface correlation of the maps.⏎All five caves occur in the Tomstown Dolomite (dip 25º E., strike N.-S.). The area itself is almost entirely surrounded by metamorphic or ign
-
is effective in controlling mosquitoes that transmit human diseases such as Aedes spp., Anopheles gambiae, Anopheles culicifacies, Bamisia tabaci, Cule<|endoftext|>The American Association of Feline Practitioners (AA
-
687|⏎|Julkaisu||JOURNAL OF LANGUAGE TEACHING AND RESEARCH|⏎|DOI - pysyväislinkit|⏎|Tila||Julkaistu - 1 syysk. 20
-
directional or reverse current relays differential relays, and distance relays.⏎According to timing characteristics the relays can be divided into instantaneous, definite time lag, inverse time lag and inverse definite minimum time (IDMT) lag relays.⏎In
-
where four major flyways converge. This area is also along a major migration route bordering the Gulf of Mexico, where many common species that winter in the Neotropics might linger until Christmas.⏎For more information on the annual count, visit
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: one who who is who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who who
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
data points, it mainly deals with unlabelled data.⏎Some applications of unsupervised machine learning techniques include: clustering (grouping coherent segments of customers,
-
, we enhance these assumptions by considering mechanically diverse layers with varying dips. We employed the Finite Element Method (FEM) to create numerical models simulating two chamber shapes: a circular form and
-
shipbuilding, aerospace, prosthetics, architectural projects and even special effects in movies. It is essentially the evolution of a merger between a graphical user interface and Non-Uniform Rational B-Splines (NURBS) geometry, and currently functions
-
designs show significant improvement over microelectronic designs in metrics such as footprint area and power consumption. These improvements are obtained from designs employing nanodevices and nanofabrication techniques that already have been demonstrated experimentally. Thus, the results presented
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
- L20:238 excites
- L23:115 excites
- L10:656 excites
- L4:879 weakly excites
- L3:105 weakly excites
- L3:838 weakly suppresses
- L3:1333 weakly excites
- L19:746 weakly excites
- L16:974 weakly suppresses
- L7:1328 weakly suppresses
- L20:139 weakly excites
- L11:3046 weakly excites
- L2:2447 weakly excites
- L10:763 weakly suppresses
- L9:1917 weakly suppresses
- L16:270 weakly excites
- L6:1991 weakly excites
- L7:1071 weakly suppresses
- L12:3041 weakly excites
- L5:78 weakly excites
- L12:69 weakly excites
- L1:2612 weakly suppresses
- L18:2681 weakly excites
- L8:2721 weakly suppresses
Feeds later layers
- L27:2511 weakly excites
- L27:541 weakly suppresses
- L26:2524 weakly excites
- L26:3024 weakly excites
- L27:2266 weakly suppresses
- L27:2529 weakly suppresses
- L27:1335 weakly excites
- L26:1224 weakly suppresses
- L26:2894 weakly excites
- L27:2893 weakly excites
- L27:1726 weakly excites
- L27:4 weakly excites
- L26:754 weakly excites
- L26:2023 weakly suppresses
- L26:444 weakly excites
- L27:2392 weakly excites
- L26:3041 weakly excites
- L27:1991 weakly excites
- L26:462 weakly excites
- L27:1051 weakly excites
- L26:1722 weakly excites
- L26:2938 weakly excites
- L26:1550 weakly suppresses
- L27:2010 weakly 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
- ete
- etes
- الداخلية
- ValueEventListener
- itempty
- ewis
- (grammarAccess
- 子弟
- öße
- akov
Suppresses
- .gg
- 죽
- อง
- IGN
- alty
- 模式
- yr
- 名义
- richt
- -channel