'be' after 'to' in infinitives
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 responds to the word 'be' when it comes after 'to' or 'designed to', as in 'not meant to be taken' or 'suitable to be read'. Turning it up or down did not clearly change the topic of what the model wrote.
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.
-
appellation of his is known in China, Târanâtha mentions nine more names: Kâla (Time), Durdarsha (Hard-to-be-seen), Durdarshakâla (Hard-to-be-seen
-
treat spinal curvature, including scoliosis and kyphosis, in our patients at Woodbury Spine.⏎Adult spinal curvature⏎Your spine is not designed to be completely straight. It has subtle front-to-back curves to help provide support and movement
-
and hotels. They hide nearly everywhere and spread easily from one place to another in backpacks, suitcases, linens, shoes, clothes, and boxes to be near a food source.⏎After feeding, they move to secluded places and hide for
-
“St. Valentines,” each clamoring to be recognized as the day’s patron saint, but here’s the most romantic story:⏎
-
their way. If you cheat, your transaction does not get forwarded.⏎I hope that explains the concept of how Bitcoin works in regards to your question. Just to be clear:⏎"Does each bitcoin has a unique identifier,
-
Add your thoughts below. You can sign in with Google, Twitter, Facebook, Yahoo! to get your story posted straight away - or just post a comment to be published later.⏎Let us help you plan a route and make a visit<|endoftext|>Pre
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 question is: This is a question that is still: This is a question that is not. I think that's the answer. So, in the case of the question is: The answer is. So, in the case of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Enterprise, were mobilized to respond to a distress call from Vulcan, which was under attack by the Narada. After the destruction of Vulcan, acting captain Spock planned to regroup with the rest of the fleet in the Laurentian system to
-
rampant poaching that takes advantage of the “arribadas” (arrival of nesting sea turtles
-
ley nests each year.⏎WildAid completed an enforcement assessment last year of three marine protected areas (MPAs) in Nicaragua’s coral corridor to address turtle poaching and other threats to the region’s biodiversity. Sea turtle eggs are considered delicacies
-
Deacons are also supposed to have a demonstrated ability to manage their households well. This may suggest they had certain administrative functions within the church.[i]⏎There may have been other areas entrusted to deacon as well. The term deacon, after
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
- L0:1183 weakly suppresses
- L0:1 weakly suppresses
- L2:1036 weakly excites
- L1:2340 weakly suppresses
- L2:2940 weakly excites
- L0:1224 weakly suppresses
- L2:483 weakly suppresses
- L0:1675 weakly suppresses
- L2:2930 weakly suppresses
- L1:92 weakly excites
- L0:1247 weakly suppresses
- L0:178 weakly suppresses
- L1:1742 weakly excites
- L1:1363 weakly excites
- L0:2522 weakly excites
- L0:1301 weakly suppresses
- L0:1256 weakly suppresses
- L2:1343 weakly suppresses
- L2:1121 weakly suppresses "for" after "hard" or "difficult"
- L2:1899 weakly excites
- L0:294 weakly suppresses
- L2:1657 weakly excites
- L0:2007 weakly suppresses 'key' after 'The'
- L2:1184 weakly excites
Feeds later layers
- L16:1096 weakly suppresses
- L19:2934 weakly suppresses
- L15:284 weakly suppresses
- L7:2777 weakly suppresses
- L11:2471 weakly excites commas in lists of six items
- L18:1749 weakly excites
- L16:879 weakly suppresses
- L6:1932 weakly suppresses
- L8:529 weakly excites
- L24:2893 weakly suppresses
- L8:2296 weakly excites
- L14:2770 weakly suppresses
- L18:1779 weakly suppresses
- L9:1027 weakly excites 'to' or 'on' after 'was' or 'were' (planned future action)
- L22:758 weakly excites
- L9:2242 weakly suppresses
- L7:2647 weakly suppresses
- L17:2307 weakly excites
- L4:3018 weakly excites
- L18:1955 weakly suppresses
- L25:3067 weakly excites
- L10:2446 weakly suppresses
- L4:2553 weakly suppresses
- L25:565 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
- shake
- hawk
- pie
- amen
- auty
- trade
- uar
- shares
- maid
- pher
Suppresses
- �能
- quires
- iveness
- erno
- ọi
- 您的
- ively
- Reconstruction
- 辙
- 万名