comma after a city in an address
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 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 comma right after a city name in an address or dateline, such as 'Swansboro, NC' or 'WASHINGTON, DC'. Turning it up or down gave text with no clear link to addresses.
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.
-
Internet’s powers of mobilization and information dissemination. But it was last summer’s huge<|endoftext|>Proceedings of the Fourth Urban Forestry Conference⏎ST. LOUIS, MISSOURI, OCTOBER 15-19, 1
-
NC 28570 and at Hammocks Beach State Park located at 1572 State Rd 1511, Swansboro, NC 28584.⏎The Coastal Federation’s Recycling for Reefs
-
L., B. Davis, et al. (2006). Does horse manure harbor invasive plants? Cal-IPC Symposium. Rohnert Park,
-
Afternoon highs of 111° at Redding, CA, 107° at Medford, OR and 97° at Miami, FL were new records for the date. At Miami, it was the third of
-
015 The University of North Carolina Press⏎116 South Boundary Street, Chapel Hill, NC 27514-3808⏎How to Order
-
, 2013⏎David H. Trinklein, 573-882-9631⏎COLUMBIA, Mo. – It
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 re or the re or re in the re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re re
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
additional physical abnormalities. LEOPARD syndrome is an autosomal dominant genetic disorder.⏎LEOPARD syndrome and Noonan syndrome are both caused by mutations in the PTPN11 and RAF1 genes.⏎The symptoms and physical characteristics associated with LE
-
place when the body enters a second phase called the endotoxin tolerance (ET) phase, where the opposite occurs.⏎Scientists examined the role of a protein called ABCF1 in white blood cells during inflammation in a mouse model of sepsis
-
aspect of the vitamin’s medicinal properties is particularly significant for young children whose skeletal structures are still in the developmental stage. For them, insufficient vitamin levels can result in rickets and
-
3 to 8 days, but small infants who have been infected may be contagious for up to 4 weeks.⏎In small infants RSV also causes mucus membrane damage and clear drainage
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
- L21:369 excites
- L21:577 suppresses
- L21:1512 suppresses
- L24:1970 suppresses
- L24:141 excites
- L19:1340 excites
- L21:1180 suppresses
- L20:661 excites
- L24:534 excites
- L21:384 excites
- L22:216 excites
- L21:682 excites
- L22:406 excites
- L16:1982 excites
- L18:1242 excites
- L21:230 suppresses
- L23:216 suppresses
- L20:1585 suppresses
- L19:1753 excites
- L20:193 suppresses
- L23:242 excites
- L22:2753 suppresses
- L24:26 suppresses
- L19:1399 suppresses
Feeds later layers
- L26:224 suppresses
- L27:22 suppresses
- L27:159 suppresses
- L27:258 excites
- L27:38 excites
- L27:19 excites
- L27:20 excites
- L26:240 suppresses
- L27:65 suppresses
- L27:6 suppresses
- L26:27 suppresses
- L27:133 weakly excites
- L27:17 weakly excites
- L27:64 weakly suppresses
- L26:155 weakly excites
- L27:316 weakly excites
- L26:92 weakly excites opening parenthesis before a gloss
- L26:300 weakly suppresses
- L27:83 weakly excites
- L27:379 weakly suppresses
- L27:468 weakly suppresses
- L26:96 weakly suppresses
- L27:248 weakly suppresses
- L26:630 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
- 透视
- �单
- лит
- 仅供参考
- perature
- 眼光
- 打造成
- ví
- stärke
- ---</
Suppresses
- b
- Th
- Sh
- In
- X
- Gen
- Sin
- K
- How
- Re