hedges like 'it turns out' and 'if any'
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. 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 that finishes a small set phrase, such as 'turns out', 'if any', 'if possible' or 'would say'. Turned up, it nudged the model toward wordy, qualified sentences, though the effect was mixed and sometimes just repeated a word.
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.
-
that lights up when we see a baby or romantic partner, as well as the brain area stimulated by money, food, or sex. Giving, as it turns out, may offer as big a reward to the giver as to the receiver.⏎Giving
-
minimum wage.⏎Eighty-seven percent of single parent households are headed by women and of those 37% are in poverty (Association for Women’s Rights and Development).
-
Why don't you be the first to send us one?⏎If Google Street View is available, the image is from the best available vantage point looking, if possible, towards the location of the monument. Where it is not available, the satellite
-
look at how animation on the Internet came to its present (and very cool) state. It pretty much all started with gifs…⏎.gif files are, it turns out, older than I am by about two years. They were introduced in 1
-
also been an alarming increase in reports of anti-trans murders and other violence especially those perpetrated against younger American transwomen of color. However, there are few, if any, reports of anti-gay or anti-trans violence perpetrated against visitors.⏎There are many
-
, "Why, now is this seemingly obscure story from the Book of Genesis not only included in the Book of Ruth, but included in what is, in my contention, the climax of the entire Book?"⏎The passage Iím referring to is a
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, the, author, in,,,ressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressressress
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
weeks of learning online, which is open to all. The course format allows users to learn at their own pace, also engaging with others taking the course if they choose. There are no exams or assessments.⏎“It was an extraordinary experience to be the
-
puritan society. In his novel,The Scarlet Letter, Nath<|endoftext|>What they learned after only seven years is worth remembering (or learning, as the case may be).⏎The United States has operated under two constitutions. The first, The Articles
-
No approach. Meier was surprised by the assumption that non-marital sex has negative psychological consequences. The physical risks — unwanted pregnancy or sexually transmitted diseases, for example — were obvious.
-
general informational purposes only and is not a substitute for professional medical advice.⏎If you think you may have a medical emergency, call your doctor or (in the United States) 911 immediately. Always seek the advice of your doctor before starting
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
- L16:650 strongly excites
- L15:1159 strongly excites ampersand in citation author lists
- L16:2905 strongly suppresses
- L15:2506 strongly suppresses
- L15:841 strongly suppresses
- L16:2626 strongly excites
- L15:2666 strongly excites
- L14:2932 strongly excites
- L17:207 strongly excites
- L15:2961 strongly suppresses
- L16:1788 excites
- L16:331 suppresses
- L16:2571 suppresses
- L15:2201 suppresses
- L15:3010 excites
- L17:1668 excites
- L16:3060 suppresses
- L17:1299 excites
- L15:2235 suppresses
- L17:975 suppresses
- L17:449 excites
- L17:1829 excites
- L16:2360 suppresses
- L17:260 suppresses commas mid-sentence before a continuation
Feeds later layers
- L19:845 strongly suppresses
- L21:344 strongly suppresses
- L24:572 strongly suppresses
- L19:682 strongly suppresses
- L26:236 strongly suppresses
- L27:86 excites
- L20:276 suppresses
- L20:756 suppresses
- L20:173 suppresses
- L27:286 suppresses
- L21:1070 suppresses
- L20:82 excites
- L25:438 suppresses
- L20:1083 suppresses
- L26:22 excites the 'res' fragment inside words
- L19:1130 suppresses
- L19:1201 suppresses
- L24:642 suppresses
- L24:6 excites
- L19:1468 suppresses
- L20:132 suppresses
- L21:301 suppresses
- L23:49 suppresses
- L24:304 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
- ANTED
- หาย
- licos
- agrams
- otive
- ulares
- encias
- ilian
- 柯
- caled
Suppresses
- ,
- ,"⏎
- .,
- .",
- **,
- [],
- !,
- ,
- ,",
- ",