'it' before 'is important to'
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. 23 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
The neuron lights up on the word 'it' when it starts a sentence after a phrase like 'Therefore,' and is followed by a phrase such as 'is important to' or 'is required'. When it was turned up, the model got stuck repeating the word 'is'.
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.
-
will have a significant impact on your health and fitness. Therefore, it is important to choose a particular school of yoga. Ashtanga yoga can be
-
textile materials in the single light source will not be adequate. To avoid such circumstances, it is required to test the colours from the fabrics in distinctive light shades with the goal
-
the dangerous gases owing to the following properties:⏎Tips for Safe Storage of Compressed Acetylene Gas Cylinders⏎Owing to the above discussed hazards, it has become important to take special care while storing and handling compressed gas cylinders of ac
-
essential to ensure that it receives sufficient stimulus for muscle growth. Not all muscles working in a multi-joint movement will necessarily produce the same muscle growth. Therefore, it may be advisable to include additional exercises that specifically target the muscles with stimulating repetitions in
-
the classification of cloud computing contracts in the Turkish and European legal frameworks.⏎1. Cloud computing and cloud computing service models⏎Prior to classifying cloud computing contracts, it is important to understand what kind of services are provided pursuant to such contracts and who
-
symptoms depends on many factors, including your dog's size and how much of the plant he ingested. Severe cases can result in death, however, so it's vital to contact your veterinarian immediately if you suspect your dog has chewed on
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: should one should not need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the candidate pool more quickly.⏎If every question halved the size of the candidate pool, then four or five closed (yes/no) questions should be all that’s required. However, Guess Who is
-
found all the ways?⏎Swap the stars with the moons, using only knights' moves (as on a⏎chess board). What is the smallest number of moves possible?⏎Can you work out how to balance this equaliser? You can put
-
can it recover low-grade heat for power generation, it can also be done using thinner, lighter devices than semi-conductors. One slim cell is all that is needed to regulate the thermal cycle, with graphic oxide applied on one side
-
to look for:⏎No special preparation is needed before the test. A chest x-ray will be performed before and after the
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
- L22:912 strongly excites
- L22:1041 suppresses
- L20:2349 excites
- L20:2994 excites
- L19:2765 excites
- L21:2956 suppresses
- L22:1588 excites
- L20:2182 suppresses
- L21:1990 excites
- L22:691 suppresses
- L19:1597 suppresses
- L21:1827 suppresses
- L20:2706 suppresses
- L18:1813 suppresses
- L18:2998 excites
- L22:1227 excites
- L17:246 excites
- L21:1199 excites
- L15:2584 suppresses
- L20:2818 suppresses
- L20:1696 weakly excites
- L21:2280 weakly excites
- L18:522 weakly suppresses
- L14:2332 weakly suppresses
Feeds later layers
- L26:109 strongly suppresses
- L27:500 excites
- L25:822 excites
- L27:804 excites
- L24:565 excites
- L24:299 excites
- L24:319 excites
- L26:1182 excites
- L26:978 weakly excites
- L24:1279 weakly excites
- L27:1256 weakly excites
- L27:1236 weakly suppresses
- L24:789 weakly suppresses
- L24:764 weakly suppresses
- L26:1276 weakly suppresses
- L25:1674 weakly excites
- L24:1690 weakly excites
- L24:2503 weakly excites
- L24:240 weakly suppresses
- L27:1123 weakly suppresses
- L25:9 weakly suppresses
- L25:1471 weakly suppresses
- L24:1584 weakly suppresses
- L27:1275 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
- is
- 是
- .is
- 是比较
- 是非常
- 是很
- Is
- is
- là
- 的是
Suppresses
- must
- 必须
- moeten
- ENCED
- deben
- 应该
- 应在
- otron
- accountId
- 应