opening parenthesis after a name or term
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. 12 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 lights up on the open bracket that comes right after a name or term, like a politician's party in brackets or an abbreviation tucked after a word. When it was turned up, the model's writing got stuck in loops and repeated phrases rather than saying anything clearly related.
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.
-
the pointers is larger, this value may be positive or negative.<|endoftext|>Think of your body like a car. Where the muscles are the shock systems and the skeleton (joints/bones) is the frame of the car. The job of the
-
the electric current,⏎- V is the voltage.⏎(Note: this is not true where the impedance is complex)⏎|This page uses Creative Commons Licensed content from Wikipedia (view authors).|<|endoftext|>Of the moon that is! The start of 2
-
morning, Bible is scheduled to testify at a Capitol Hill hearing before the Senate Agriculture Committee, which is considering the carbon-farming legislation. Introduced by senators Mike Braun (R-Indiana) and Debbie Stabenow (D-Michigan), the
-
equivalent ... )⏎The Amplitude response looks somehow like this :⏎Fig.4: Measured Amplitude Response S21⏎You will notice, that the maximum (fs) will move while the minimum (fp) remains constant.⏎With the short
-
. On the same day, the Consumer Financial Protection Bureau (CFPB) released a study reviewing financial education research priorities for America’s youth.⏎Congressmen Bill Foster (D-IL) and French Hill (R-AR) both introduced H.
-
Following is the sequence:⏎- Chinta (Anxiety)⏎- Asha (Hope)⏎- Shraddha (Rever
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 want to be able to do something, so I need to think about something. I want to be able to do something... I want to be able to do something... I want to be able to do something... I want to be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
San Tan Allergy & Asthma can help determine whether your (or your child’s) symptoms may be related to a food allergy or not and can
-
913|⏎|Red List Category & Criteria:||Endangered B1ab(i,ii,iii,v) ver 3.1|⏎|Reviewer(s):||Butchart, S. & Symes, A.|⏎|Contrib
-
title Augusta.⏎Titus, in full Titus Vespasianus Augustus, original name Titus Flavius Vespasianus, (born Dec. 30, 39 ce—died Sept.
-
usoleums while reflecting on life.⏎Maybe the attraction of visiting an abandoned city is a similar feeling — the emptiness and loneliness, wondering where these people (or their offspring) might be now.⏎Dir Journal
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
- L10:66 suppresses opening parenthesis after a citation or term
- L4:64 excites
- L8:354 weakly excites opening parenthesis before a citation year
- L9:1390 weakly suppresses
- L3:57 weakly excites
- L5:330 weakly excites opening parenthesis after a noun
- L1:2324 weakly suppresses
- L9:2010 weakly suppresses
- L1:822 weakly excites
- L9:1055 weakly suppresses
- L1:1857 weakly suppresses
- L10:2391 weakly excites "not" in "if not" contrasts
- L7:104 weakly excites
- L6:915 weakly excites
- L10:2556 weakly excites
- L3:218 weakly suppresses
- L6:1968 weakly excites
- L11:2158 weakly excites
- L4:85 weakly suppresses
- L4:3040 weakly excites
- L10:1041 weakly suppresses opening parenthesis before phone or code
- L11:50 weakly excites
- L7:13 weakly suppresses 'the' after 'member of' or 'one of'
- L2:2057 weakly suppresses
Feeds later layers
- L16:2289 excites 'or' linking two alternatives
- L14:2156 excites opening parenthesis after a term
- L16:574 suppresses
- L14:2276 excites opening bracket after a term or number
- L22:144 excites
- L16:1065 weakly excites
- L13:958 weakly excites
- L13:820 weakly suppresses comma after an introductory phrase
- L17:2274 weakly excites
- L14:1265 weakly excites
- L17:882 weakly suppresses
- L22:973 weakly excites comma after a full date
- L13:1108 weakly excites
- L19:1864 weakly excites
- L19:2576 weakly excites
- L23:1619 weakly excites
- L18:2544 weakly excites
- L15:2486 weakly excites bracket or parenthesis inside a quote
- L24:1598 weakly excites
- L17:1577 weakly suppresses
- L21:1807 weakly excites
- L17:1708 weakly excites
- L13:933 weakly suppresses
- L16:2425 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
- част
- icc
- 송
- hop
- ño
- pired
- -sizing
- roy
- CLOSE
- 人造
Suppresses
- editors
- FRING
- istrates
- ofil
- legislators
- Beirut