dash starting a new list item
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. 7 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 responds to the dash or bullet mark that begins a new item in a list, especially a list of questions or points. Turning it up pushed the text toward more questions and quiz-style lines.
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.
-
corium (dermal soft tissue); third phalanx (bone P3)⏎How quickly does sole and wall grow, and what is the normal length?⏎- sole: 4mm/month⏎- wall: 5-6mm/month
-
10 commandments from God?⏎- What are we told to do in the fifth commandment?⏎HONOR OUR FATHER AND OUR MOTHER⏎- What was the name of Moses' brother?⏎
-
bad? Why?⏎- How did the children feel when the giant told them to leave?⏎Watch the video and then talk about the discussion points with an adult.⏎o Who did Tia want to<|endoftext|>| Quote #1⏎Richard's first mention
-
This prophet was swallowed by a fish.⏎- What animal was often used for transportation in the Bible?⏎CAMEL or DONKEY (either answer is correct)⏎- Who was the Son of God?⏎- What was the name of Jesus' mother
-
Balance Sheet (Assets, Liabilities & Equity)⏎• Understanding Cash Flows (Operating, Investing & Financing)⏎Analysing & Interpreting Financial Statements⏎• Understanding Vertical Analysis⏎• Understanding Ratios (Performance, Liquidity & Leverage
-
daily basis.⏎- Countries from Africa, Asia-Pacific and the European Union too have expressed their support for the Bali Declaration.⏎About Minamata Convention on Mercury⏎- The Minamata Convention on Mercury is the most recent global agreement on environment and
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 can't. I'm just a number. I am a number. I don't know. I don't know. So, I can't. I can't. So, I can't. I can’t. I can't
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
results demonstrate that silkworm-BEVS is a promising alternative strategy to produce the functional and thermostable reverse transcriptase.⏎All Science Journal Classification (ASJC) codes⏎- Insect Science<|endoftext|>An account of the religion of
-
0,10 250,190 160,210" style="fill:lime;stroke:yellow;stroke-width:1"/> </svg>⏎Here is what the above code looks like in HTML
-
-layer coated, triple-layer coated and graded-index-layer coated samples was performed and compared with experimentally derived films. Good agreement between model and experiment was observed.⏎ASJC Scopus subject areas⏎- Electronic, Optical and Magnetic Materials⏎- Ceramics
-
diverse set of controls, as well as in a discontinuity analysis on observations near the borders between anglophone and francophone countries.⏎- Colonial rule⏎ASJC Scopus subject areas⏎- Economics and Econometrics⏎- Organizational Behavior and
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
- L6:164 suppresses
- L5:329 excites dash starting a new list item
- L4:453 excites
- L6:606 suppresses
- L5:103 weakly suppresses
- L5:89 weakly suppresses
- L6:172 weakly suppresses
- L2:209 weakly excites 'also' after 'not only ... but'
- L4:1074 weakly excites
- L5:877 weakly excites
- L4:25 weakly excites
- L0:1163 weakly excites
- L4:15 weakly suppresses
- L2:2112 weakly suppresses
- L4:139 weakly suppresses
- L4:19 weakly suppresses
- L6:13 weakly excites
- L2:111 weakly suppresses
- L5:775 weakly suppresses
- L4:97 weakly suppresses
- L4:288 weakly suppresses
- L2:375 weakly suppresses
- L5:2904 weakly suppresses
- L6:1113 weakly excites
Feeds later layers
- L10:1676 weakly excites
- L8:1226 weakly excites
- L12:1544 weakly excites
- L11:729 weakly excites
- L8:1036 weakly excites
- L8:2812 weakly excites
- L17:1854 weakly excites
- L8:186 weakly suppresses
- L13:2437 weakly excites
- L19:1757 weakly excites
- L26:1888 weakly suppresses
- L21:2948 weakly excites
- L10:681 weakly suppresses
- L13:371 weakly excites
- L15:2450 weakly excites
- L22:90 weakly suppresses
- L17:806 weakly excites
- L14:1946 weakly excites
- L9:229 weakly excites
- L10:332 weakly excites
- L25:1694 weakly suppresses 'and' joining paired names or fields
- L24:2550 weakly excites
- L11:1022 weakly excites
- L15:1050 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
- Cove
- amarin
- cribed
- adelphia
- riad
- hait
- itate
- すでに
- case
- licity
Suppresses
- *</
- goog
- ollo
- 火烧
- nehmen
- burnt
- entlich
- ーン
- ulg
- *</