opening parenthesis after a measurement
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. 13 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 opening bracket that comes right after a measurement, like the "(" in "80 metres (295 ft)", where the same amount is repeated in another unit. Turning it up made the model's writing wander into rambling, unrelated sentences, so the effect on ordinary text was not clear.
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.
-
edited for style and length.⏎Cite This Page:<|endoftext|>Briers, Mississippi facts for kids⏎Quick facts for kids⏎|Elevation||62 ft (19 m)|⏎|Time zone||UTC-6 (Central (C
-
since apps have made their way onto most electronic devices, you might think there would be dozens of worthwhile products available on the topic designed for students. Think again.⏎Sure, there are plenty of apps devoted to November’s election, there just aren’t
-
Norman cathedral. It is the seat of the Bishop of Coutances and Avranches, previously the Bishops of Coutances. Standing 80 metres (295 ft) tall, it dominates the town and can be seen from
-
(66 feet) and weighed in at roughly 50 tonnes (55 tons), or equal to around eight African bush elephants.⏎The dinosaur is
-
centimetres). Temperatures are high, and they change little during the year. The coolest month has an average temperature no lower than 64 degrees F (18 degrees C). The temperature difference between day and night is greater than the
-
perianth segments. There can be as many as half a dozen flowers on a stalk. The flowering stalk may stand up to 3 ft (1 m) tall. The sword shaped leaves are mostly basal, in two ranks
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 new development in the country's new development of the country's new policy of the country's new success of the country's successful in the country's success of the country's overall success of the country's success to achieve the overall success of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
-Brava⏎platform with 500 m bathymetric contours and 200m topographic contours on Fogo. Dotted areas mark⏎seismically active regions in the channel between Fogo and Brava (se
-
in⏎fructooligosaccharides. Levrat et al. (1991) demonstrated that even low⏎levels of inulin are effective in decreasing plasma cholesterol in rats.⏎DESIGN:
-
OMG). OMG has called for information on object-oriented⏎methodologies, that might create a rigorous software modelling language. Many⏎industry leaders have responded in earnest to help create the standard.⏎See also: ST
-
. Ordained in 1868, his success as a “pastor,⏎preacher, liberal theological teacher and social reformer” led to his⏎appointment as head of Scots Church, Collins Street (1875-8
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
- L11:2928 excites
- L12:2666 suppresses
- L10:553 suppresses list-style essay phrasing, weak
- L11:2971 excites
- L12:852 weakly excites
- L10:157 weakly suppresses end of document and web boilerplate
- L8:13 weakly suppresses
- L5:209 weakly suppresses
- L6:472 weakly excites
- L0:964 weakly suppresses
- L9:511 weakly suppresses
- L3:2510 weakly suppresses the word 'operation' and its forms
- L9:99 weakly suppresses
- L2:61 weakly suppresses
- L10:58 weakly excites
- L10:1715 weakly suppresses
- L9:1560 weakly excites
- L7:1188 weakly suppresses
- L11:163 weakly suppresses
- L12:2553 weakly suppresses
- L1:307 weakly suppresses
- L10:2049 weakly excites
- L11:784 weakly suppresses
- L7:276 weakly suppresses
Feeds later layers
- L17:814 suppresses
- L14:385 excites
- L14:194 suppresses
- L15:1229 excites
- L15:1785 excites
- L15:2635 suppresses
- L17:496 suppresses
- L15:2979 weakly excites
- L14:209 weakly excites
- L18:847 weakly suppresses
- L14:1080 weakly suppresses
- L14:564 weakly suppresses
- L25:456 weakly suppresses
- L14:2351 weakly suppresses
- L20:1190 weakly excites
- L21:2026 weakly suppresses
- L14:2869 weakly suppresses
- L26:288 weakly excites
- L26:373 weakly suppresses
- L17:2384 weakly excites
- L15:1570 weakly excites
- L17:1718 weakly excites
- L14:905 weakly suppresses
- L17:933 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
- ⏎⏎
- 流水
- An
- RNA
- :⏎⏎
- Höhe
- lòng
- 高的
- isi
- VALID
Suppresses
- ",(
- �建
- ',(
- _QMARK
- ncias
- NSBundle
- ?(
- reak
- #__
- ')</