unclear: word before a stated value or definition
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. 16 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 is hard to pin down. It tends to light up on small words like 'is', ':' or '-' that come right before the point where a value, definition or explanation is about to follow. When it was turned up, the model's text drifted into repeated lists about presidents and cars, which doesn't clearly tell us much.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
recognizing alternative interpretations in their writing. (A sample story and four figures of data are included.) (RS)⏎Publication Type: Speeches/Meeting Papers; Reports - Research⏎Education Level: N/A⏎Authoring Institution: N/A<|endoftext|>A
-
). The state of cell is determined by a set of rules, which all cells must obey. These rules are applied to every cell simultaneously. The only player input is the initial world state.⏎Conway's Game of Life is played by creating an
-
to other parts of the world to keep it up. Given all these pressures, we should be smart about how we use our resources — and one of our resources is soil.”⏎Bradley is collecting data from researchers who’ve already measured global crop output
-
electrical engineering positions of various certification and licensure exams. Examples and case studies are included.⏎File type: Downloadable PDF file (Note: Print book not included. To purchase print copy
-
boycott of Uruguay.⏎Format: Knockouts, semi-finals, final⏎Golden boot: Leonidas (Brazil)⏎
-
.⏎How Do I Set Up A Website Using Iis?⏎You can access the Internet Information Services (IIS) Manager by selecting Control Panel > Administrative Tools. Select Add Website from the Connections panel, expand your host tree, right-click on Sites
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 to be alone in my work. I have to be alone in my work. I have to be alone in my work. I have to be alone in my work. I have to be alone in my work. I have to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
), association mining (customer analytics), and latent variable models (factor analysis).⏎Semi-supervised machine learning⏎As the name may suggest, this model combines elements of supervised and unsupervised learning. It works by using both labeled and unlabeled
-
. Pat skin dry and immediately apply a layer of moisturizer while skin is still slightly damp. Choose one that contains ingredients known to be effective for dry skin such as hyaluronic acid, glycerin, alpha-hydroxy acids or
-
steopenia / Osteoporosis⏎Osteopenia is a term used to describe bone density that is somewhat lower than normal, but not low enough to be diagnosed as osteoporosis. A person with osteopenia is
-
infrared radiometer indirectly measures the temperature of a very thin layer (about 10 micrometres thick) or skin of the ocean (leading to the phrase skin temperature) representing the top millimeter; a thermometer attached to an already fast
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
- L15:2033 suppresses
- L10:1748 excites
- L15:1438 suppresses
- L14:1881 excites
- L15:2787 suppresses
- L11:1651 suppresses
- L15:1588 excites
- L13:287 suppresses
- L15:1497 suppresses
- L15:1176 excites
- L14:497 suppresses
- L12:2163 excites
- L14:2990 excites
- L14:2532 weakly excites
- L14:2989 weakly suppresses
- L14:962 weakly suppresses
- L14:376 weakly excites
- L14:2904 weakly excites
- L14:2235 weakly suppresses
- L15:1092 weakly excites
- L8:1097 weakly suppresses
- L14:2193 weakly suppresses
- L14:2244 weakly suppresses
- L7:2093 weakly suppresses
Feeds later layers
- L17:2057 excites
- L17:1151 excites
- L17:867 weakly suppresses unclear: punctuation before an explanation
- L19:1412 weakly suppresses
- L19:666 weakly excites
- L18:1088 weakly excites 'and' or 'the' joining two parallel items
- L18:989 weakly excites
- L23:968 weakly excites
- L18:2361 weakly suppresses
- L19:924 weakly excites
- L18:2204 weakly excites
- L19:411 weakly suppresses
- L19:354 weakly excites
- L18:801 weakly excites
- L17:1877 weakly excites
- L17:2861 weakly excites
- L20:714 weakly suppresses
- L22:2649 weakly suppresses
- L19:573 weakly excites
- L18:2173 weakly excites
- L17:2711 weakly excites
- L17:2602 weakly excites
- L20:731 weakly excites
- L17:663 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
- 的答案
- 短线
- bec
- 外交
- 不过是
- FN
- became
- 括
- Sec
- ’an
Suppresses
- ELLOW
- ildo
- upported
- OnInit
- ыва
- ancellable
- zwarte
- isposable
- ellig
- <![