the verb 'guide' or 'navigate'
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. 3 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 'guide' (or 'navigate') when it is used as a verb meaning to lead or direct someone, as in 'use X-ray to guide the catheter'. Its output points toward predicting that word next. Turning it up did not make the writing about guiding.
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.
-
enroll<|endoftext|>- Choosing a selection results in a full page refresh.⏎- Press the space key then arrow keys to make a selection.⏎- Use left/right arrows to navigate the slideshow or swipe
-
success in the transition to adulthood by preparing and mapping out services for them as early as 14 years old. Below are some suggestions and resources to help you navigate the transition to adulthood for your child or teen with a developmental disability in the South
-
hopes the helicopter will eventually carry tools around the planet and guide astronauts when they arrive.⏎The rover launches in July 2020,
-
the insertion of the catheter. Once the catheter has been inserted into the arteries, our cardiologist will use an X-ray screen to guide the catheter to the blocked coronary artery. Once there, an expandable balloon
-
), and some highlights are in Roberson (1980, 1986, 1993). The Kaufman book — Field Guide to Advanced Birding — should be in everyone's library as an immediate reference.
-
⏎In this chapter, I outline the theories related to the core concepts of my research topic. The information I present in the conceptual framework will direct my study and guide my interpretation of data. The concepts based on my theoretical framework include values, values
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 the same as I have been doing." So, if you think about the same as you have done, then the same as you have been. Now, the same as you have been. So, if you
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, as well as the symbol of a globalized world. In Bogotá in 2007, the General Chapter of the Dominican friars brought forward an analysis of the ambiguous double face of today’s globalized world: “From
-
19th. As a minor shareholder, with just enough shares to ask questions, Greenpeace has brought forward a motion calling on Statoil to withdraw its investments from the tar sands.⏎
-
in textual study of Madhva Vaisnava and other sastric works.⏎Nandinagari script is the western variety of the archaic Nagari script of northern India. Nandinagari is also found in the inscriptions
-
using the disciplinary methods passed down by his father.⏎“I have always believed that the way my parents disciplined me has
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
- L2:2429 weakly suppresses
- L0:1912 weakly excites
- L2:527 weakly excites
- L2:1657 weakly excites
- L2:111 weakly suppresses
- L2:1891 weakly excites
- L0:1665 weakly excites
- L2:300 weakly excites
- L2:2999 weakly excites
- L0:1472 weakly excites
- L0:1594 weakly excites
- L1:249 weakly suppresses "to" after "help" plus object
- L1:1244 weakly suppresses "of" after "board" or "off"
- L2:2001 weakly excites
- L2:2167 weakly excites
- L1:61 weakly excites
- L1:1491 weakly suppresses 'for' after 'known' or 'renowned'
- L2:1953 weakly excites
- L2:2768 weakly excites
- L2:1250 weakly excites
- L1:491 weakly excites
- L2:2302 weakly suppresses
- L2:2464 weakly suppresses
- L2:75 weakly suppresses
Feeds later layers
- L4:1758 excites
- L26:2820 excites
- L22:2435 excites
- L18:1069 excites
- L12:2070 weakly excites
- L10:1460 weakly suppresses
- L20:1444 weakly suppresses
- L27:2284 weakly excites
- L26:2597 weakly excites
- L16:2168 weakly suppresses
- L20:1490 weakly suppresses
- L25:2505 weakly excites
- L8:2642 weakly excites
- L4:1406 weakly excites
- L25:1990 weakly excites
- L25:595 weakly suppresses
- L20:2310 weakly excites
- L18:2027 weakly suppresses
- L6:2160 weakly suppresses
- L23:2824 weakly excites
- L12:1757 weakly suppresses
- L4:1553 weakly excites 'to' after 'for someone' (purpose infinitive)
- L5:2141 weakly excites
- L7:2594 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
- -guide
- 的带领下
- Guides
- guide
- _guide
- guides
- guidance
- Guide
- /utility
- 指引
Suppresses
- amous
- iliated
- @$
- omon
- barrel
- _$
- res
- -cl
- hos
- def