'called' after 'so'
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 'called' when it follows 'so', as in 'so-called'. Turning it up did not change the replies in any way that matches this pattern.
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.
-
can just concentrate on learning how to make machines obey digital commands. The engineering grunt comes after the intellectual workout.⏎You might think it strange that the makers of so-called educational computers are so busy trying to maintain their market share that they failed to see
-
through a buy-an-acre program. It is not uncommon for loggers, oil and gas companies, cattle ranchers, and miners to illegally extract resources from so-called “protected” areas.⏎Traditional forest communities are often the best stewards of the
-
, are of some specified fixed number of bits. For example, Java collections use 32-bit hash codes.⏎In cryptography applications, we often need a so-called secure hash function. Secure hash
-
location and length of horizontal transfers (HTs) are known for very few species. The usage of short oligonucleotides in a sequence (the so-called genomic signature) has been shown to be species-specific even in
-
space?" The variety of answers illustrates the function of fundamental analogies or models in physics as well as the role of so-called unobservable entities. Forces and Fields presents an in-depth look at the science of
-
light, the researchers have shown that the signaling pathways, which are activated by the receptors play a crucial role in controlling movement.⏎Unlike conventional methods, with the so-called optogenetics, the researchers were able to target one cell type. "We
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 problem of a girl with a number, and the number of the problem. What did you think? It's a problem about a girl. But here is a problem. Let's go through this. So, etc. So, and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
uncovered that Patagonia was filling jackets using suppliers who plucked down from birds that were still alive and force-fed so their livers could be used as foie gras. After the Patagonia incident, many other brands underwent investigation and uncovered
-
rebuilding the economy. A letter from King Bela IV refers to Jews in the cities of Pressburg (Bratislava), Senica, Trnava, Pezinok, Nitra, and Trencin. In 1
-
, 1971, p. 37; Boyce, 1977, p. 42).⏎In other Middle
-
Ewan Boyce and later the property of her husband Captain William Boyce<|endoftext|>A Point of View: Making friends the shared currency way⏎- 2
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
- L1:92 weakly excites
- L2:1206 weakly suppresses
- L1:2801 weakly excites
- L2:1692 weakly suppresses
- L1:2056 weakly suppresses
- L1:2054 weakly excites
- L0:2456 weakly excites
- L1:2679 weakly suppresses
- L2:506 weakly suppresses
- L1:1979 weakly excites
- L1:87 weakly suppresses
- L2:1955 weakly suppresses
- L0:980 weakly suppresses
- L2:1899 weakly excites
- L1:1963 weakly suppresses
- L2:2801 weakly suppresses
- L1:2938 weakly suppresses
- L1:571 weakly excites
- L2:493 weakly suppresses
- L0:1259 weakly suppresses
- L2:2397 weakly excites
- L0:2062 weakly excites
- L1:97 weakly excites full stop ending a heading or label
- L2:1765 weakly suppresses
Feeds later layers
- L13:1666 weakly excites
- L5:2799 weakly excites
- L18:977 weakly excites
- L9:968 weakly excites
- L15:1581 weakly excites
- L12:1141 weakly suppresses
- L4:577 weakly suppresses
- L19:1208 weakly suppresses
- L23:1484 weakly excites
- L12:811 weakly excites
- L9:2691 weakly excites
- L4:2471 weakly suppresses
- L14:675 weakly suppresses
- L10:2446 weakly excites
- L17:294 weakly suppresses
- L4:1125 weakly suppresses
- L7:1745 weakly suppresses
- L24:779 weakly suppresses
- L5:2289 weakly suppresses
- L19:298 weakly excites
- L8:394 weakly excites
- L4:146 weakly excites
- L15:58 weakly excites
- L6:1425 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
- thin
- 闩
- spécial
- ector
- darn
- üc
- 技术支持
- ncpy
- 主力
- şa
Suppresses
- LLU
- чив
- anker
- anken
- ante
- явление
- Kan
- ente
- kan
- ética