'corners' in Four Corners
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. 2 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 strongly on the word 'corners', especially in the phrase 'Four Corners' (the US region where four states meet). Turning it up did not make the writing mention corners, so the effect on output is unclear.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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 have, be a good example to them.⏎#9. Keep at It.⏎Talking about mone<|endoftext|>Next: My Tiles are Cracking in the Corners⏎My tiles are wearing out⏎Tiles are well known for their durability and wear
-
Utes have natural gas, the Navajo Nation has coal, which has for decades been extracted to feed various large-scale power plants, including in the Four Corners. But that revenue has come under threat by tightening air-quality regulations. At the
-
Service. "The weed population appears to be increasing exponentially in certain areas of our state."⏎Invasive weeds that choke out native vegetation are creeping into the Four Corners area from the north, south, east and west. This "slow-motion explosion
-
Diffuse knapweed can spread when dried seed heads attach to vehicles, people or animals.⏎Other knapweed varieties also are major problems in the Four Corners area. Russian knapweed is the most common, overtaking fields and empty
-
monsoon rains to return to the Four Corners and Western Slope in July. But there’s still a danger in the mountains
-
ainer: What can the oil companies do to stave off internal pipe corrosion? Very little. It would be far<|endoftext|>DLTK's Crafts for Kids⏎Cone Santa Paper Craft⏎Set Santa out on a coffee table or the mantle as a
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 thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
2. Subclinical diabetes⏎Subclinical diabetes is understood when the patient is completely asymptomatic and fasting blood glucose level is usually normal but the post prandial level (glucose levels after two hours of meal) is frequently elevated.⏎
-
14). To elucidate the drought responses of these two tree species to a long-term and repeated drought in monoculture and mixed plantings, the “Kranzberg Roof Project” (KROOF, kroof.wzw.tum
-
maintaining shrubs: heading back, renewal, rejuvenation, and shearing. For information on pruning specific shrubs, download the following list.⏎TIME TO PRUNE⏎A general rule-of-thumb is that shrubs that bear flower buds on the
-
mama putro mama bhrAtA mama vatso.atha gaurmama |⏎vR^ikairvyApAditA hyevaM krandanti sma gR^ihe gR^ihe ||2-
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
- L1:1110 weakly excites
- L1:5 weakly excites
- L0:178 weakly suppresses
- L1:2240 weakly excites
- L0:1224 weakly suppresses
- L1:1 weakly suppresses
- L1:1099 weakly excites
- L0:1301 weakly suppresses
- L1:23 weakly suppresses
- L1:99 weakly excites
- L1:18 weakly excites
- L1:1960 weakly excites
- L1:2118 weakly suppresses
- L1:331 weakly suppresses
- L1:4 weakly suppresses
- L1:538 weakly suppresses
- L1:1454 weakly suppresses
- L1:2638 weakly suppresses
- L0:497 weakly suppresses
- L1:1510 weakly suppresses
- L0:114 weakly suppresses
- L0:1632 weakly suppresses
- L0:1403 weakly excites
- L1:134 weakly excites
Feeds later layers
- L19:1605 weakly excites
- L12:1347 weakly suppresses
- L7:160 weakly suppresses
- L17:1892 weakly excites
- L24:1352 weakly excites
- L24:300 weakly suppresses
- L4:2944 weakly suppresses
- L3:2601 weakly excites
- L7:1081 weakly excites
- L13:1534 weakly excites
- L15:1984 weakly suppresses
- L9:2668 weakly excites
- L17:1670 weakly excites
- L16:2250 weakly suppresses
- L11:991 weakly excites
- L13:602 weakly suppresses
- L10:2060 weakly suppresses
- L26:1193 weakly excites
- L15:1792 weakly excites
- L4:2016 weakly excites
- L3:2112 weakly excites
- L10:2918 weakly excites
- L17:1605 weakly suppresses
- L14:2934 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
- �
- face
- o
- treffen
- atsby
- 理论
- 面
- styl
- 扪
- face
Suppresses
- |RF
- lined
- lining
- Bail
- sendo
- |r
- \|
- каж
- |R
- oogle