'of' in 'this type of'
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 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 responds to the word 'of' when it comes right after 'type', as in 'this type of failure' or 'this type of game'. Turning it up did not produce text that clearly reflected this pattern.
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.
-
for this type of post-disaster performance, Waffle House has become well prepared for Business Continuity
-
outcome. While some areas of their specialty may overlap, each member has a distinct role to play in the patient’s recovery.⏎Here are a few:⏎This type of rehabilitation deals with
-
stop it, a valve needs to be closed but the situation is too dangerous to send humans in. At the moment, no type of robot can do this type of job and that is what Kim wants to create - a robotic first-responder -
-
Zinc is also susceptible to liquid metal cracking, but only when the temperatures are much higher (750F range). The most recently reported case of this type of failure (August 1998) was on a fleet of school buses
-
shoulders are intricately decorated with epaulets bearing silver oak leaves on a blue field, indicating the Infantry branch of the United States' Army. This type of dress coat would likely have been worn for portraits and formal occasions.<|endoftext|>Much can
-
from a Chinese speaker.⏎The demo also relies on a new method for processing speech patters, using a system called Deep Neural Networks. The company says this type of machine learning is "patterned after human brain behavior," and means that the researchers
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: - - - - - - - - - - - - - - - - - - - - - - - -
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
house⏎• Add 3D furnishings such as sofas, chairs, beds, and more⏎• Perform 3D photorealistic rendering using the latest rendering engine from AutoCAD⏎• Make an animated walk-through video from your 3D
-
grow in complexity with time.⏎The global pattern to follow is MVC and the layering:⏎(View & Controller)⏎|View||HTML templates usually generated using a template engine|⏎|Input Controller||Filter and validate data|⏎||Application Controller||Application flow
-
Tok video-sharing app is a psychographic predictive machine learning engine, whose algorithm can intelligently catalog individual profiles through reactions to the seemingly innocuous
-
sub-matrix format.⏎The data file format for nmrPipe files is the serial format. It is possible, by
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
- L1:254 suppresses
- L1:492 weakly excites
- L1:386 weakly suppresses
- L1:925 weakly suppresses
- L1:266 weakly suppresses
- L1:1857 weakly suppresses
- L0:1640 weakly excites
- L1:2667 weakly suppresses
- L1:1529 weakly excites
- L0:176 weakly excites
- L1:166 weakly suppresses
- L0:2024 weakly suppresses
- L1:1769 weakly excites
- L1:92 weakly excites
- L1:2683 weakly excites
- L0:25 weakly excites
- L0:1301 weakly suppresses
- L1:173 weakly excites
- L1:87 weakly suppresses
- L0:317 weakly excites
- L1:1043 weakly excites
- L0:178 weakly suppresses
- L0:174 weakly suppresses
- L1:718 weakly suppresses
Feeds later layers
- L20:119 weakly excites
- L11:2733 weakly excites
- L10:104 weakly excites
- L5:1731 weakly suppresses
- L21:1294 weakly excites
- L7:116 weakly excites
- L6:3006 weakly excites
- L12:1922 weakly suppresses sentence-ending periods in explanations
- L6:502 weakly excites
- L7:887 weakly suppresses
- L5:679 weakly excites
- L7:3 weakly suppresses
- L8:1984 weakly excites
- L6:1906 weakly excites
- L10:2172 weakly excites
- L6:1555 weakly suppresses
- L15:908 weakly suppresses
- L10:465 weakly suppresses
- L4:855 weakly suppresses
- L7:420 weakly excites
- L3:576 weakly suppresses
- L10:2848 weakly excites
- L15:814 weakly suppresses
- L3:995 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
- нию
- ража
- יים
- нии
- Tues
- елеф
- irty
- .$.
- 斗志
- POINTER
Suppresses
- local
- rogate
- -local
- local
- 您的
- 强力
- كر
- 你的
- 顾
- Local