'the' right after 'On' in 'On the other hand'
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. 0 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 reacts to the word 'the' when it comes straight after 'On', as in 'On the other hand' or 'On the contrary'. Turning it up made the model get stuck repeating words like 'the number of', which does not match the trigger.
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.
-
MR file formats does not make for very exciting⏎reading. On the other hand many NMR spectroscopists have experienced the⏎frustration
-
olds plants every four to six weeks. On the contrary, you can also use organic gardeners to work in plenty of compost at
-
transformed these unlikely men, most of them elitist, militaristic, and fiercely nationalistic, into martyrs to a universal ideal?⏎The resisters in On the Road to the Wolf’s Lair are not the singular souls doomed to failure by the
-
default on a loan or not. For this, you have a target variable i.e. ‘defaulting on loan’, which is a supervised learning problem. On the other hand, customer segmentation in a market has no target variable which makes it an
-
success I have enjoyed as a man,” Peterson said in a statement.⏎Among those coming to Peterson’s defense was NBA Hall of Famer Charles Barkley. On the CBS pregame show The NFL Today, Barkley recalled childhood beatings. “
-
On the other side, they⏎should write one sentence about Abraham Lincoln's life. After
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 thinking about what is the main thing for the next project. I have been thinking about the project I am thinking about. I have been thinking about the project. I have been thinking about the project. I have been thinking about
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
5 eroding from a bank at approximately⏎25 cm below the current terrain⏎surface (see in-situ photo below).⏎Its front surface,⏎face down in the photo, is carved
-
and perhaps as quickly as 30 days - an incredible accomplishment for what was thought to be a loosely organized society consisting of small, widely scattered bands of foragers.⏎"What's extraordinary about these findings is that it provides some of the first evidence
-
which migrate to the dermis and trigger the release of cytokines in particular the tumor necrosis factor-alpha which causes skin inflammation which is coupled by increased production
-
inside them…” (Inferno 8, line, 73). This is meant to warn the narrator that the city dwellers are hostile and inhuman to strangers. They even further refer
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
None: this is the first layer.
Feeds later layers
- L16:792 excites "the" inside a prepositional phrase
- L26:1941 weakly excites
- L9:24 weakly excites "the" before a named organisation or title
- L3:719 weakly excites digits inside numbers and statistics
- L26:1080 weakly excites
- L16:419 weakly excites
- L18:641 weakly excites
- L23:288 weakly suppresses
- L6:419 weakly excites
- L1:1566 weakly suppresses
- L18:2243 weakly suppresses
- L23:1406 weakly excites
- L13:477 weakly excites
- L22:1318 weakly suppresses
- L2:2838 weakly suppresses 'Head Start' and 'follow up'
- L27:6 weakly excites
- L2:766 weakly suppresses
- L21:614 weakly excites
- L14:1169 weakly excites
- L11:2201 weakly excites
- L23:2214 weakly suppresses
- L17:30 weakly excites
- L17:1652 weakly suppresses
- L20:1246 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
- above
- main
- opposite
- critical
- attached
- around
- amount
- company
- feet
- objective
Suppresses
- unnel
- üssen
- ossa
- utra
- 下来的
- 回事
- 不高
- 走过
- 素质教育
- villa