'the' in mid-sentence noun phrases
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. 19 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' when it appears in the middle of a sentence and a noun or adjective is expected next. The steering results do not show a clear theme, so it is hard to say what the neuron does beyond that.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
, Wegener's groundbreaking theory was turned upside down.⏎Earthquakes are not only terrible natural disasters, they also offer a view inside the Earth. It was the geophysicists Wadati and Benioff, who in 1
-
and you’re all set.⏎At one time, all engine blocks were cast iron. Today, they’re aluminum. Be it iron or aluminum, the process is the same, minus the melting point.⏎What is Sintered Metal?⏎Before any
-
provide the same amount of heat to the product as the boiling water processes. Just because pressure is used to decrease the process time, the canning process is not the same as one to destroy spores of Clostridium botulinum as
-
or rag. These low-tech lamps smoked and gave off faint, flickering light, similar to candles. Candles and devices using this oil-burning technique were the main means used to light homes, employed in candelabras, wall scon
-
OCT uses light waves for imaging. The shorter wavelength of light empowers this technology to generate high-resolution, cross-sectional images with zero delays.⏎Time Domain technology was the first method that was used in OCT technology. Time Domain technology uses a low-co
-
iniae cause severe intestinal infections. Studies using Yersinia pseudotuberculosis as a model organism aim to elucidate the infection mechanisms of these...⏎Researchers from the
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 to be very careful, and I have to be very careful, and I have to be very careful, I have to, I have to, and I have to. (I have been writing this, but I have written,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Guide⏎The Production Process⏎- Domestic Resource and Production Projections⏎- Outlook for Natural Gas Prices⏎- Economic Benefits and Implications⏎- Public Health and the Environment⏎States Take Action: The Balancing Act⏎State Policy Actions⏎-
-
-interaction. The students are also encouraged to seek out and take advantage of other opportunities that will further their expertise: computer-related part-time employment, internships, the Computer Science club, field trips, invited speakers, LAN parties :), etc.
-
JSTOR JSTOR is part of ITHAKA, a not-for-profit organization helping the academic community use digital technologies to preserve the scholarly record and to advance research and teaching in
-
justice policies.. All of these plans are informed by the Federal GND — they are explicit plans to address not only the climate crisis but also racial justice, the housing crisis, the pandemic and a Just<|endoftext|>Wheels, like other load-bearing components
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
- L18:171 strongly excites
- L15:888 strongly excites
- L18:1359 strongly excites
- L16:652 strongly suppresses
- L11:1168 strongly excites
- L9:662 strongly suppresses
- L18:641 suppresses
- L16:1594 excites
- L10:120 suppresses
- L18:1913 suppresses
- L18:1225 suppresses
- L18:216 excites
- L15:1888 excites
- L18:2630 suppresses
- L10:1377 suppresses
- L17:30 suppresses
- L10:321 suppresses
- L10:1644 excites
- L9:227 excites
- L4:73 suppresses
- L3:344 excites
- L1:740 suppresses
- L9:2198 excites
- L16:792 excites "the" inside a prepositional phrase
Feeds later layers
- L25:377 excites
- L24:109 suppresses
- L20:1047 excites
- L21:675 excites
- L22:260 suppresses
- L24:153 excites
- L20:1989 suppresses
- L20:240 suppresses
- L23:218 suppresses
- L27:60 suppresses
- L22:1081 suppresses
- L25:335 excites
- L26:111 excites
- L22:211 suppresses
- L25:390 suppresses
- L20:2117 excites
- L24:360 suppresses
- L23:79 excites
- L24:434 excites
- L25:812 excites
- L20:889 excites
- L26:552 excites
- L20:2092 excites
- L20:932 suppresses 'of' after a count or quantity
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
- decisions
- predictions
- transactions
- reactions
- instances
- discussions
- responses
- examples
- seminars
- figures
Suppresses
- epad
- 印象深刻
- ��
- exporter
- ductor
- odus
- rometer
- atum
- ocese
- classpath