closing bracket ending an aside
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. 14 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 closing bracket that finishes a short extra phrase in brackets, such as "(accounting phase)" or "(orange cap)". It was weaker on the closing bracket after source names like "(Nanowerk News)". Turning it up did not give a clearly related theme.
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.
-
1957 and also participated in the introduction of the Euro (EUR) in a two-phased approach in 1999 (accounting phase) and 2002 (monetary phase) to replace the Belgian Franc
-
, we used transcranial magnetic stimulation (TMS) to disrupt neural activity in the RTPJ transiently before moral judgment (experiment 1, offline stimulation) and during moral judgment (experiment 2, online stimulation). In both experiments,
-
wine (to the left) and similar representations of Lower Egypt (to the right), On their heads the former
-
3,000 genes over 3Gb of DNA). Genes were found corresponding to genes for energy metabolism, both aerobic (used by the juveniles) and anaerobic (used by the adults). While the genes coding for proteins
-
: it runs back and forth like a courier (a cursor is an ancient Roman runner or messenger), trying to deliver the goods before the race (cursus) is over. Hard
-
sessions (majalis) in meeting places known as zawiyas, khanqahs or tek
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: it's been a lot of time in the last month) of) time) time) time) time) time) ) time) time) time) ) time) time) ) time) ) time) ) time) ) time)
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
lotz Reuters⏎12:17 p.m. EDT, August 8, 2012⏎PASADENA, California (Reuters) – NASA’s newly landed Mars science rover Curiosity snapped the first color image of
-
things of interest⏎Posted: September 29, 2006⏎Magnetic nanobridge brings two worlds together⏎(Nanowerk News) A gadolinium layer of no more than one nanometer in thickness is capable
-
of interest⏎Posted: June 27, 2009⏎Nanoscale holes provide speed boost for diagnostic tests⏎(Nanowerk News) Microfluidic devices, essentially miniaturized chemical laboratories etched into material similar
-
02, 2012⏎Repairing bones with nanomaterials⏎(Nanowerk News) Around half a million people in the EU need treatment for serious skeletal defects every year
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
- L8:926 suppresses
- L3:33 excites
- L7:879 excites
- L2:1384 suppresses
- L10:1883 excites closing parenthesis after a number or unit
- L4:516 weakly suppresses
- L8:2208 weakly excites
- L5:378 weakly suppresses
- L5:696 weakly excites
- L4:542 weakly suppresses
- L0:522 weakly excites
- L8:2599 weakly excites
- L10:522 weakly excites
- L2:979 weakly excites
- L5:1079 weakly suppresses
- L11:35 weakly suppresses
- L10:304 weakly excites
- L1:1749 weakly excites
- L10:33 weakly suppresses
- L8:334 weakly suppresses
- L10:791 weakly suppresses
- L3:891 weakly suppresses commas closing a list item
- L8:1341 weakly excites
- L4:602 weakly suppresses
Feeds later layers
- L15:616 excites closing parenthesis ending an aside
- L15:1429 excites
- L15:1554 suppresses
- L15:421 excites
- L18:142 excites
- L15:2298 excites
- L17:506 excites
- L17:2157 excites
- L15:324 excites
- L15:2220 excites
- L19:1381 suppresses
- L16:611 suppresses
- L17:1824 weakly excites
- L16:665 weakly excites
- L15:2906 weakly suppresses
- L20:281 weakly excites
- L16:1155 weakly excites
- L15:764 weakly excites
- L16:1179 weakly suppresses
- L16:413 weakly excites
- L17:335 weakly excites
- L17:1541 weakly excites
- L19:601 weakly suppresses
- L15:1625 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
- ensible
- imize
- rist
- ponsible
- ATAL
- isions
- ⊢
- fluence
- gary
- 腈
Suppresses
- past
- *)⏎⏎
- ')</
- 发言人
- 岛
- phas
- platz
- .);⏎
- %");⏎
- över