answer labels and tag questions
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. 17 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 most to the word 'Answer' or a colon right before a reply, as in study worksheets. It is strongly pushed the other way by tag questions such as 'isn't it?' or 'doesn't he?'. Turning it down made the model produce a stream of 'right?' tag questions.
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.
-
are parallel, perpendicular or neither?⏎Parallel lines are the lines that never intersect and perpendicular lines are the lines that intersect at 90 degrees.⏎Answer: Lines with the same slope are parallel and if the slope of one line is the
-
is the total probability of finding a particle in a one-dimensional box in level n = 3 between x = 0 and x = L/ 6?⏎Answer to relevant QuestionsWhich has the lowest (ground-state) energy, an electron trapped
-
70%.⏎Answer: B. Humidity levels should be kept below 60 percent, ideally between
-
she could read prior to enrolling in the course. She obtained the following five number summary:⏎Use this information to draw a boxplot of the reading speed.⏎Answer to relevant QuestionsThe following
-
done.⏎Answer: True. There are hundreds of thousands of dust mites in our homes,
-
's disease, and how do you differentiate between regular memory loss and memory loss associated with Alzheimer's disease?⏎Answer: So, the memory changes of Alzheimer's disease represent a change for that person
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: right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right, right,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
inventor, Stairs. It's funny how you walked the stairs so many times and had no idea about who invented them and why they are called that, isn't it?<|endoftext|>Conventional, high-yield farming could have a better shot at
-
(TLB)⏎Jesus delayed on purpose so God’s glory would be shown and others would believe in him.⏎Sometimes Jesus does the same with our requests, doesn’t he?⏎He delays. We know he can. No doubt. We pray he
-
is simpler for you to collaborate with, and will certainly use up much less of your time than searching for individual worksheets.⏎This sounds like a horrible lot, does not it! A great collection of preschool worksheets should cover every one of this, and
-
craft (or lolly) sticks and rubber bands.⏎An amazing STEM project for future scientists – sling shot LEGO car! It even sounds so much fun, don’t you think?⏎Setting up fine motor practice really doesn’t have to be hard or
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
- L15:2412 suppresses
- L16:1059 excites
- L14:441 weakly excites
- L8:1298 weakly excites
- L16:741 weakly suppresses
- L4:2258 weakly suppresses
- L8:2275 weakly suppresses
- L9:846 weakly excites
- L15:1388 weakly excites
- L1:1983 weakly excites
- L15:348 weakly suppresses
- L12:44 weakly suppresses
- L4:2564 weakly suppresses
- L4:1922 weakly suppresses
- L1:1950 weakly suppresses
- L16:376 weakly excites 'or' joining alternatives
- L3:2668 weakly excites
- L9:2524 weakly excites
- L14:2026 weakly suppresses
- L0:918 weakly excites
- L2:760 weakly excites
- L7:2377 weakly excites
- L0:1077 weakly excites
- L13:417 weakly excites
Feeds later layers
- L20:2335 strongly suppresses comma ending an 'if' clause
- L24:2138 suppresses comma before 'right?' tag
- L18:594 excites
- L20:2730 suppresses
- L21:1538 suppresses
- L19:835 suppresses
- L20:1922 suppresses
- L24:402 suppresses
- L20:1957 weakly suppresses
- L26:613 weakly suppresses
- L21:1362 weakly suppresses
- L25:541 weakly suppresses
- L21:1484 weakly suppresses comma after a sentence subject
- L22:387 weakly excites
- L19:2793 weakly suppresses
- L21:1044 weakly suppresses
- L20:459 weakly excites
- L18:1912 weakly excites
- L20:1838 weakly excites
- L24:2161 weakly suppresses
- L24:2257 weakly suppresses
- L26:1937 weakly suppresses
- L19:670 weakly excites
- L20:1371 weakly suppresses comma inside a question
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
- _ly
- blockbuster
- bled
- tridge
- 骘
- 感动
- _mode
- OMATIC
- erry
- ONY
Suppresses
- 吗
- 吧
- ?
- right
- 嗎
- ?⏎
- 了吗
- 么
- 右
- ですよね