'then' in step-by-step sequences
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. 11 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 'then' when a text describes steps in order, such as 'start with X, and then Y'. Turning it up did not produce a clear change in the stories the model wrote.
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.
-
is a severe type of foodborne illness that causes paralysis by blocking the motor nerve endings; the paralysis progresses downward, usually starting with the eyes and face, and then moving to the throat, chest and extremities<|endoftext|>The Peckham Experiment⏎
-
ing design and nozzle selection of musical fountain .⏎For the overall design of the creative fountain, the basic requirements of the environment and atmosphere should be analyzed first, and then the various waterscape forms should be analyzed, different combinations should be listed, and
-
take on the world!⏎Talk, Sing and Point⏎Babies learn language from the moment they are born. To a newborn baby, speech is just sound. Then, day by day, they learn that sounds have meaning. Every time you talk
-
to our overall health. But if you’re not a person who has always practiced this habit how do you begin? Start with making one healthy choice per meal and then one entire healthy mindful meal per week. More on mindfulness tomorrow.⏎Depending on your
-
that doesn't mean you can't teach your 2-year-old to respect others. Begin with the basics (saying "please" and "thank you"), then try these six other strategies.<|endoftext|>This archived Web page remains online for reference,
-
» Have a routine. It’s easier to calm down before bed when kids do the same things in the same order. Start with bath time and teeth brushing, then settle in for a bedtime story or section of a chapter book.⏎» Say no
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 writing about the idea that the community is a place where we use language and tools to make sense of our lives and the idea that we use language to make sense of our lives. This is the case every day, but the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in a manger.” 13 And suddenly there was with the angel a multitude of the heavenly host praising God and saying,⏎14 “Glory to God in the highest,⏎and on earth peace among those with whom he is pleased!”
-
research in the field supporting a link between vocabulary and comprehension.”⏎On a scale of 0-500, fourth-graders scored 224 and eighth-graders scored 2
-
movement quantum of light behave as a corpuscular (no charge). b) Using its own inner impulse / intrinsic angular momentum ( h* = h / 2pi ) circle - particle rotates around its axis. In such movement particle has charge,
-
as “peace on earth.”⏎9 And an angel of the Lord appeared to them, and the glory of the Lord shone around them, and they were filled with great fear. 10 And the angel said to them, “Fear not
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
- L5:540 excites
- L4:592 excites
- L7:946 weakly suppresses
- L6:1652 weakly excites
- L9:1667 weakly excites "before" after a do-this-first advice clause
- L8:1477 weakly excites
- L9:2181 weakly excites
- L7:2426 weakly excites
- L8:1514 weakly excites 'turn' in 'which in turn'
- L9:1477 weakly excites
- L9:1748 weakly excites
- L9:1780 weakly suppresses
- L5:2958 weakly suppresses
- L9:1830 weakly suppresses
- L9:3018 weakly suppresses
- L6:891 weakly excites
- L6:2934 weakly suppresses
- L7:2896 weakly excites
- L0:433 weakly suppresses
- L1:2808 weakly suppresses
- L5:2380 weakly suppresses
- L5:2367 weakly suppresses
- L6:2445 weakly suppresses
- L2:1846 weakly suppresses
Feeds later layers
- L16:2387 excites
- L15:961 suppresses
- L12:2481 excites
- L17:2973 excites
- L15:193 suppresses
- L16:2591 weakly excites
- L14:2375 weakly excites
- L13:490 weakly excites
- L19:1863 weakly excites
- L16:937 weakly excites
- L12:1272 weakly excites
- L19:1687 weakly excites
- L13:1085 weakly excites
- L19:3058 weakly excites
- L18:1700 weakly excites 'so' in 'and so on'
- L17:1318 weakly suppresses
- L23:429 weakly excites
- L19:2997 weakly excites
- L22:2419 weakly excites
- L15:456 weakly excites 'and' between two verbs
- L16:2949 weakly excites
- L13:1724 weakly excites
- L13:1435 weakly excites
- L19:2349 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
- ucceeded
- mieux
- urther
- [next
- athed
- .Fl
- 埽
- ️
- .Stage
- .Back
Suppresses
- คำตอบ
- nock
- kho
- gamle
- ogh
- ilyn
- Vox
- cesso
- ty
- ?><