"the" in "to the next"
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. 27 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 "the" in phrases like "from one level to the next" or "until the next". Turning it down makes the model repeat the word "next" over and over.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
example⏎84 rounded to the nearest ten would be 80.⏎Numbers that end in a digit of 5⏎or more should be rounded up to the next
-
children between the ages of two and five. Color is also used to emphasize the different layers lodged inside the toy.⏎In order to successfully progress from one level to the next, the child needs to find the correct orientation for each piece that will enable
-
100 to get the percent. This method is also useful for finding the exact value of the number. Once you have the answer, you can move onto the next step in interpreting the result.⏎The first step in finding the percentage of a
-
point. For example, if the gradient magnitude is 2.5 and the learning rate is 0.01, then the gradient descent algorithm will pick the next point 0.025 away from the previous point.⏎If you
-
explore side-branches, they make mistakes. The “right answer” – the proof, the demonstration – simply sets a mile-marker between one phase of discovery and the next.⏎In our Math Circles, young people invent and discover their shared way
-
. Once your child runs out of cash in an envelope, they have met their budget for that pay period and are unable to spend any more in that category until the next pay period. Creating a challenge encourages your kids to ask themselves important questions:
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: next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next next
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
tiny wasps in the Encarsia genus. Integrated pest management specialists with the University of California, Davis recommend avoiding the use of pesticides containing diazinon, carbaryl, pyrethroids and imidacloprid when caring for
-
country, there has been a rapid decline of both native plant and animal species. Some introduced plants have become invasive, taking over where wild plants once thrived. Planting native plants in your backyard is a step towards preserving your own
-
⏎- Maximize use of lubricants in service⏎- Reduce maintenance and lubrication costs⏎- Extend equipment component life⏎Tests for oil condition include:⏎- Viscosity measures the resistance of a fluid to flow, is the most important property
-
plant⏎Brambles (rubus spp.) — blackberries/dewberries⏎Grape (vitis spp.) — muscadine⏎Greenbriar (smilax
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
- L23:1339 strongly suppresses
- L24:1027 strongly suppresses
- L26:1578 strongly excites
- L23:2609 strongly excites
- L20:2152 strongly suppresses
- L22:2156 strongly suppresses
- L19:2997 strongly suppresses
- L23:2678 suppresses
- L21:2426 suppresses
- L20:1858 suppresses
- L19:1736 suppresses
- L18:3047 suppresses
- L8:14 excites
- L18:2132 suppresses
- L22:2671 excites
- L22:1183 suppresses
- L26:1564 excites
- L22:201 excites
- L8:7 excites
- L22:2338 suppresses
- L16:2387 suppresses
- L25:2579 excites
- L22:1953 excites
- L5:110 suppresses
Feeds later layers
None: this is the last layer.
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
- eken
- altimore
- itudes
- obody
- uros
- DET
- инфек
- 檔
- juana
- êu
Suppresses
- next
- next
- Next
- Next
- (next
- →next
- _next
- .next
- [next
- ,next