'such' after 'no' or '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. 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 fires on the word 'such' when it follows phrases like 'no such', 'any such' or 'such an', which are common in negative or quantity statements. Turning it up produced rambling, generic text without a clear link to the idea.
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.
-
9’s negative impacts on the cardiovascular system highlights a great need for identifying COVID-19 patients at risk for heart problems, the researchers say. However, no such predictive capabilities currently exist.⏎“This project will provide clinicians with early warning signs and ensure
-
a prince’s brain knows what his mouth’s doing?” Bertie: “You’re obviously not well acquainted with many royal princes.”⏎No one could have imagined any such dialogue involving Archduke Otto von Habsburg, who died on July
-
by inserting a verse about the Exodus into the third paragraph of the Shema], thus ensuring that we never forget this important historical event and its attendant lessons, no such obligation exists (according to most Halachic authorities, with the possible exception of
-
Piet has dreamed of proving that he is a skater as brave and strong as Pim Mulier -- but he had never imagined that his test would fall under such dangerous circumstances. Louise Borden's moving text captures all the tension, excitement,
-
ing, it is likely to subside into quiescence as soon as its object is fulfilled by the supramental being's appearance. But there is no such stress in reality: the idea of human progress itself is very probably an illusion,
-
⏎The Museu Civico in Sassoferrato is fortunate to hold such an important and rare relic of the Palaiologan Renaissance, the final period
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 have been thinking about lately, and I have been thinking about lately. So I want to make a difference, and I want to make a difference. I have been thinking about lately, and I have been thinking about lately. So I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
comes as the maverick Australian speech therapist, Lionel Logue, is just getting to know His Royal Highness Prince Albert, the stammering Duke of York:⏎Logue
-
Diddle Diddle, All Around the Mulberry Bush, I’m a Little Tea Pot, Hickety Pickety Bumble Bee, The Grand Old Duke of York, Ride a Cock-horse to Banbury Cross, Pat-a-cake,
-
her daughters’ safety and arrange appropriate marriages for them and so she permitted them to leave sanctuary and attend his court.⏎Henry VII married her eldest daughter, Elizabeth of York
-
to ascertain the nature of humanity and of God and to facilitate the experience of the presence of divine love and wisdom in the world.⏎Sufism, or Taṣawwuf ( الْتَّصَوُّف;
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
- L2:2293 suppresses
- L1:1056 weakly excites
- L2:122 weakly excites
- L2:1404 weakly excites
- L0:2929 weakly suppresses
- L2:1036 weakly excites
- L1:1200 weakly suppresses
- L2:79 weakly excites
- L2:934 weakly suppresses
- L0:1862 weakly suppresses
- L1:1091 weakly suppresses
- L0:1115 weakly suppresses
- L0:1472 weakly excites
- L2:1134 weakly suppresses
- L2:2227 weakly suppresses
- L0:246 weakly excites
- L0:1700 weakly suppresses
- L1:1862 weakly excites
- L2:2346 weakly excites
- L2:2424 weakly excites
- L2:2285 weakly excites
- L2:2254 weakly suppresses
- L0:2833 weakly excites
- L2:1497 weakly excites
Feeds later layers
- L4:1498 excites
- L4:2688 excites
- L5:1364 excites
- L4:468 excites "that" after "than" or "like"
- L17:2988 excites
- L6:1892 excites
- L4:653 excites "that" after "such" or "such a way"
- L5:2308 weakly excites
- L11:2331 weakly excites 'so' before much or many
- L11:1890 weakly excites
- L4:1873 weakly excites
- L6:1458 weakly excites
- L7:2434 weakly excites
- L18:2248 weakly suppresses
- L11:229 weakly excites
- L13:2092 weakly excites
- L5:2175 weakly excites
- L10:2918 weakly suppresses
- L22:2159 weakly suppresses
- L7:2380 weakly excites
- L11:1039 weakly suppresses
- L5:2567 weakly excites
- L4:1955 weakly suppresses
- L5:1372 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
- ied
- deployment
- severity
- ielding
- Hispan
- CALLBACK
- ields
- .Callback
- 椅子
- ิง
Suppresses
- acf
- awl
- aos
- apolis
- ôm
- омер
- adder
- ooth
- uell
- abox