words leading into a planned action
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. 13 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 lights up on words that set up a purpose or a plan, like 'help teachers [plan]', 'prepare for [rollback]' or 'so they can [accurately] judge'. It fires where the text is about to say what someone will do. Turning it up gave rambling text that did not clearly continue that theme.
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 crucial in his effort of making learning meaningful. Moreover, knowing how students learn the English language could help English teachers plan activities suited them. This study generally determined the English language learning strategies of the grade
-
Sunday night as Ernesto began to head northwest from Haiti towards Cuba.⏎“It would take a relatively significant change from the current forecast to prevent us from going into rollback preparations,” says launch integration manager LeRoy Cain.⏎Space shuttle officials had planned to decide
-
develop a definitive diagnosis, which allows the dental professional to rule out other possible causes of the pain (periodontal disease, caries
-
, doctors need to know if you are taking these products so they can accurately judge whether particular symptoms are related to a medical illness or a side effect from an
-
consequence of a risk assessment might be included in tender documents so that designers, contractors, subcontractors and suppliers can take them into account when pricing their tenders and planning their work. Ultimately, it may, where relevant, be included in the health
-
their students understand the content. Rather than testing what students think they know about the topic, formative assessments allow you to determine how well students comprehend content and guide instruction based on that information.⏎🤔 Challenge: Limited Interaction⏎The lack of interaction between
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 had a new rule after the first 100000 after I had done this, I had done this...? This, I had...? I had done this...? after I had...? This, I had
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
gluten per day.⏎What Have Other Studies Shown?⏎Overall, studies have delivered inconsistent results in attempting to determine whether early gluten consumption increases the likelihood of developing celiac disease. Previously conducted studies have found that there is no connection between early consumption of
-
as possible in my image) my equation was 600/18=33 seconds.⏎With an exposure of 33 seconds I had to have a tripod and a cable release, also with an exposure of 33 seconds
-
see if they later developed celiac disease.⏎Additionally, it is important to note that while the study found a relationship between eating more gluten early in life and celiac disease, this relationship is correlational, not causal. This means that the study
-
of Healthwise, Incorporated.<|endoftext|>New research is suggesting that children who eat more foods with gluten before the age of two have a greater risk of developing celiac disease if they are genetically predisposed to the condition.⏎This study provides some insight into
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
- L11:1350 suppresses
- L12:1524 excites
- L12:2094 excites
- L7:104 excites
- L12:2394 suppresses
- L11:1508 excites
- L12:2113 excites
- L12:2274 excites
- L11:1367 excites
- L7:912 suppresses
- L12:732 suppresses
- L10:946 suppresses
- L12:1463 excites
- L12:170 excites
- L10:1447 excites
- L11:1234 excites
- L12:2064 suppresses
- L12:833 suppresses
- L11:2469 suppresses
- L8:840 suppresses
- L12:2194 suppresses
- L12:1466 weakly suppresses
- L7:1860 weakly suppresses
- L12:386 weakly suppresses
Feeds later layers
- L14:3019 excites
- L14:2259 excites
- L15:2564 excites
- L14:2324 excites
- L14:515 suppresses
- L14:2323 excites
- L15:516 suppresses
- L14:2363 excites
- L14:1679 excites
- L14:1574 suppresses
- L15:3001 excites
- L14:603 suppresses end-of-document token
- L14:582 excites
- L15:1383 excites
- L14:2601 excites
- L14:2486 excites
- L14:2863 excites
- L14:2875 excites
- L14:2451 excites
- L18:2975 weakly excites
- L15:2562 weakly excites
- L21:391 weakly suppresses
- L14:457 weakly excites
- L14:96 weakly suppresses
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
- izzle
- accordingly
- getReference
- Depends
- ickness
- depending
- depends
- depended
- ере
- grew
Suppresses
- Bytes
- unate
- sheets
- nung
- 斗
- mäßig
- num
- ながら
- .rand
- ?v