the verb 'visit' before a website
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. 9 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 fires on the word 'visit' (and similar words like 'at' or 'page') when a text tells the reader to go to a website, as in 'please visit' or 'more information online at'. When it was turned up, the model started writing web addresses and links.
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.
-
will be available online at the Space Sciences Website at: http://www.perseidslive.com⏎For an electronic version of this advisory or more information, visit Marshall’s News Center Website at:http://www.msfc.nasa.gov/news⏎The
-
23, 2016⏎Current as of: May 23, 2016⏎To learn more about Healthwise, visit Healthwise.org.⏎© 1995-2016 Health
-
19 pandemic, this year’s ceremony was unable to take place in-person. To learn more about this year’s awardees and virtual commemoration, please visit https://
-
,5,28-2013.7.14, and view the complete TEKS-aligned Newspaper Collages lesson plan and others, please visit our Educator Resources page.<|endoftext|>For a long time in history
-
affects everyone. Let’s all be clean, green, environmental machines!⏎Check out the list of Earth Month events below, or for a complete list of events, visit www.pace.edu/dyson/earthmonth.⏎An Environmentally Friendly Future
-
it.⏎Volume (Year): 42 (2011)⏎Issue (Month): 2 (June)⏎|Contact details of provider:|| Web page: http://www.tandfonline.com/VECE20|⏎|
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 is to find the number of times a certain number of bits was changed... So I have a binary code of 8 bits, which is 0 1 2, and so on. I have to find the number of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
about a hormone called insulin. Depending on the type of diabetes you have, your body either can’t make enough insulin to keep blood sugar levels down (type 1), or can’t effectively use the little insulin that it does produce (type
-
compared with children not exposed to gestational diabetes, according to study results published in Diabetes & Metabolic Syndrome: Clinical Research & Reviews.⏎The results indicated that improving lifestyle habits among children exposed
-
-like organisms, playing a key ecological role in coastal ecosystems: support of food web, coastal protection of erosion, bioremediation by removal of nitrogen or phosphate
-
vic acids, increased ketosis,⏎and hypoglycemia. Treatment consists of continuous glucose supplementation⏎to avoid gluconeogenesis.
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
- L6:2240 excites
- L4:2944 suppresses
- L8:1134 excites
- L7:2259 suppresses
- L3:2296 suppresses
- L8:87 suppresses
- L4:1774 weakly excites
- L6:1724 weakly suppresses
- L5:1661 weakly excites
- L6:1355 weakly excites
- L5:2406 weakly suppresses
- L8:10 weakly excites
- L7:2233 weakly excites
- L8:1372 weakly suppresses
- L8:439 weakly suppresses
- L8:979 weakly excites
- L8:1787 weakly excites
- L8:1609 weakly excites
- L5:2113 weakly excites
- L7:2699 weakly suppresses
- L5:1189 weakly suppresses
- L1:1877 weakly suppresses
- L6:870 weakly suppresses
- L4:1920 weakly suppresses
Feeds later layers
- L15:2943 excites
- L19:1391 excites
- L10:439 excites
- L22:1354 excites
- L17:1646 excites
- L11:3004 excites
- L23:2398 excites
- L17:1534 excites
- L20:2063 excites
- L13:399 weakly excites
- L14:1122 weakly excites
- L12:773 weakly excites
- L10:1989 weakly suppresses
- L11:2415 weakly excites
- L23:274 weakly excites web browsers and online security text
- L20:806 weakly excites
- L10:2170 weakly suppresses
- L21:2312 weakly excites
- L10:2100 weakly suppresses
- L24:1425 weakly excites
- L10:1411 weakly suppresses
- L22:2005 weakly suppresses
- L24:2380 weakly excites
- L18:749 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
- 网址
- :url
- :http
- url
- url
- websites
- Homepage
- �
- URL
- ://
Suppresses
- .Orientation
- wyją
- apsed
- 鞘
- stras
- EventType
- ords
- ItemId
- itan
- utenant