document ends and record boundaries
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. 14 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
The neuron reacts at the seams between separate documents, such as the end-of-text marker or a line break just before a new record header. The strongest cases come from catalogue entries and text that stops abruptly. Turning it up did not produce a clear theme in the writing.
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.
-
O’Neil was declared a conservation area.⏎Neil burn Drive was created.⏎Canmore place was erected.<|endoftext|>ERIC Number: ED379961⏎Record Type: RIE⏎Publication Date: 19
-
⏎- Other factors (e.g. extremes in temperatures, sunlight, tight-fitting clothing, etc.)<|endoftext|>ERIC Number: ED348648⏎Record Type: Non-Journal⏎Publication Date: 1992-Apr
-
banks, insurance companies, utility companies, and other institutions, and they pass on those costs to customers.⏎A foundation-supported initiative allows Rwandan farmers to access markets<|endoftext|>The small rufous-sided pygmy tyrant is a member of the
-
. The polyps are generally removed during a hysteroscopy because the doctor has the best possible view of them during such a procedure.⏎Some women may experience<|endoftext|>Sandrine Ceurstemont, video producer⏎How does an octopus locate
-
he doubted it would have much impact. “There’s no promise of funding,” Turner said.<|endoftext|>This article touches on cultural, linguistic, and technical issues that arise in video recording
-
about it<|endoftext|>Press Release 07-030⏎Survey Reveals Family Ties
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 the the first to have been the first to have been the themselves first and then I have been the first of themselves, so I have been the first of themselves... and so on... I have been the first of themselves,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included in the page “GNU Free Documentation License”.⏎The copyright and license notices on
-
, access via your institution.⏎Buy single article⏎Instant access to the full article PDF.⏎Price includes VAT (USA)⏎Tax calculation
-
did so far:⏎Originally Posted by MrFujin⏎Enter a 5-charcter string: hgout
-
, please share it through the social media buttons.⏎When Mary arrived and saw Jesus, she fell at his feet and said, “Lord, if only you had been here, my brother would not have died.”⏎“Where have you put him?”
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
- L11:623 weakly excites
- L2:2493 weakly excites
- L5:1462 weakly suppresses
- L8:950 weakly excites
- L4:1862 weakly suppresses
- L9:979 weakly suppresses
- L2:1991 weakly excites
- L3:2345 weakly suppresses
- L12:2617 weakly excites
- L9:81 weakly suppresses
- L6:1256 weakly suppresses
- L5:415 weakly suppresses
- L13:987 weakly excites
- L13:2057 weakly excites
- L10:2830 weakly suppresses
- L9:2290 weakly suppresses
- L7:743 weakly excites
- L10:213 weakly suppresses
- L6:2513 weakly suppresses
- L7:1631 weakly suppresses
- L4:1519 weakly suppresses
- L10:507 weakly suppresses
- L4:21 weakly excites
- L6:2082 weakly suppresses
Feeds later layers
- L15:1605 suppresses
- L19:2210 excites
- L16:2306 weakly excites
- L26:223 weakly suppresses
- L21:2718 weakly excites
- L27:1347 weakly excites
- L25:300 weakly suppresses
- L25:949 weakly excites
- L19:1701 weakly excites
- L17:2944 weakly excites
- L16:1310 weakly excites
- L15:1199 weakly suppresses
- L26:373 weakly suppresses
- L24:1578 weakly suppresses
- L19:1703 weakly suppresses
- L20:166 weakly suppresses
- L27:785 weakly excites
- L15:1567 weakly suppresses
- L19:2977 weakly excites
- L27:1442 weakly suppresses
- L17:230 weakly suppresses
- L21:1460 weakly excites
- L17:2316 weakly excites
- L15:2449 weakly suppresses line breaks mid-sentence
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
- chi
- abwe
- pear
- pired
- iri
- Küche
- حتياج
- ซอ
- _SUPPLY
- elli
Suppresses
- OMET
- -------⏎
- ilent
- ___⏎⏎
- >All
- ==========⏎
- __________________⏎⏎
- =========⏎
- kneeling
- ---------------⏎