unclear: first letter of a name fragment
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. 27 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 reacts to a short first piece of a capitalised name or word, like the O in Oss or the N in Ndebele. The pattern is loose, and turning it up or down only produced repeated, unrelated words.
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.
-
995 and 2014, World Bank says. Wealth remains unequal, and some countries are falling behind.⏎El Banco Mundial analiza la riqueza de los países, tomando en cuenta no solo el PIB,
-
professor and psychologist at NTNU.⏎This research is published in the Journal of Developmental & Behavioral Pediatrics.⏎Previous news in sleep:<|endoftext|>In the Dutch city of Oss, 60 miles southeast of Amsterdam, there's a highway named N
-
insurance. About 44% of these people were white, 33% were Hispanic, 15% were black, 5% were Asian/Native Hawaiian or Pacific Islander, and 3% were of another race,
-
. In 1787, more than 100 years later, this disaster was described in the historical book "Het Schrickelijcke Oordeel
-
a common food source for many Native American tribes including but not limited to the Iroquois, Algonquin, Ojibwa, and Chippewa. It was dried for a winter
-
) and Dashkat Hakaayuʔ “Whitebread” (who told 19% of the stories) https://archive.org/details/traditionsofcadd00dorsrich⏎Explore more about the Caddo history
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: Redaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredaredared
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
called radius and apply the relevant formula to calculate and return the volume of a sphere based on the given radius.⏎- Your second subroutine should be called getVolumeOfCuboid(), take three parameters called width, length and height and apply the relevant
-
certain times, all the digits are⏎consecutive. How many times like this are there between midnight⏎and 7 a.m.?⏎These practical challenges are all about making a 'tray' and covering it with paper.⏎My cousin was
-
a liquid brew that is consumed in a ceremonial context.⏎Ayahuasca is made from two plant substances, the woody bark of the vine Banisteriopsis caapi and the leaves of Psychotria viridis. Together, these plant sources
-
-Tape measure -Plywood -Circular saw -Ruler or a straight edge -Stapler -Paint or a sealant -Cordless drill -1 inch hole saw -Paintbrush -Safety glasses⏎-
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
- L24:346 excites
- L24:375 suppresses
- L21:160 suppresses
- L22:245 weakly suppresses
- L23:2981 weakly suppresses
- L20:550 weakly excites
- L18:1703 weakly excites
- L21:237 weakly suppresses
- L16:248 weakly excites
- L18:569 weakly excites
- L20:151 weakly excites
- L4:681 weakly suppresses
- L23:322 weakly suppresses
- L22:530 weakly suppresses
- L17:367 weakly excites
- L23:268 weakly excites
- L18:247 weakly suppresses
- L24:105 weakly suppresses
- L22:700 weakly suppresses
- L0:546 weakly excites
- L18:1472 weakly suppresses
- L20:1091 weakly suppresses
- L8:2470 weakly suppresses
- L3:7 weakly excites
Feeds later layers
None: this is the last layer.
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
- đích
- rack
- 金字塔
- l
- iasm
- /stdc
- r
- 找了
- behalf
- endforeach
Suppresses
- ared
- 🚩
- adamente
- mü
- .bp
- 诸侯
- [out
- ingly
- .LayoutStyle
- ney