unclear: names and 're' word starts
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. 25 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 pattern is unclear. It fires on some names, such as Yuri Gagarin, and on 'Re' or 're' at the start of words like 'reimagining' and 'rely', but these examples do not share a clear theme. Turning it up gave unrelated text about numbers and meetings.
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.
-
6, vol.25, n.1, pp. 44-69. ISSN 1980-4369. http://dx.doi.org/10.1590/S01
-
American History: Selected Documents Celebrating the Manuscript Division’s First 100 Years features drafts of Langston Hughes’ poem, “The Ballad of Booker T.” In this 1941 poem,<|endoftext|>Whenever the stench
-
11 was a victory for the U.S. in the Cold War Space Race with the Soviet Union who, in 1961, had sent Yuri Gagarin to the Cosmos to become the first human in outer space and the
-
2014, Whales as marine ecosystem engineers, Front Ecol Environ, http://www.joeroman.com/wordpress/wp-content/uploads/201
-
WHO’s 13th General Program of Work target of ‘1 billion lives made healthier’ by 2023 and the global Education 2030 Agenda coordinated by UNESCO.⏎The global standards provide a resource for education systems to
-
told Sputnik.⏎“We will always remember the names of the first daredevils who made the first flight into space and the first round the world trip – Yuri Gagarin and Ferdinand Magellan. Without remembering the past, we will not
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 just found a new kind of cappardard on the market. It looks just like a cappardard but harder, harder, harder hard, hard, hard hard hard hard hard hard hard hard hard hard hard hard hard hard
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
other artistic production where sign language can be deeply explore and fully understand. Ryan Commerson, producer and host of the movie “Media, Power & Ideology: Re-Presenting D-E-A-D”, states that all the preconceptions societies
-
- 45 drops<|endoftext|>BROOKLYN BRIDGE FOREST⏎Our Brooklyn Bridge Forest submission was the winner of Van Alen Institute’s competition ‘Reimagining Brooklyn Bridge’, ‘which challenged participants to rethink the iconic Brooklyn Bridge walk
-
top of the bee habitat is garden sage. The Lehigh Valley’s many pollinators—hummingbirds, bees and butterflies, to name a few—rely on these plants for precious nutrients.⏎These gardens are small and easy to maintain
-
few years later<|endoftext|>Hand-coloured lithograph entitled; 'Dampfmaschinen Pferd worauf man in einer stunde von Paris nach Petersburg reiten kann' ('Steam engine horse on which one can
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
- L17:2796 suppresses
- L4:101 excites
- L3:515 weakly excites
- L15:1659 weakly suppresses
- L23:1143 weakly excites
- L2:2773 weakly excites
- L5:543 weakly excites
- L22:2239 weakly suppresses
- L22:148 weakly suppresses
- L23:903 weakly excites
- L20:1387 weakly excites
- L16:390 weakly excites
- L13:262 weakly suppresses
- L11:1997 weakly suppresses
- L18:2313 weakly excites
- L15:618 weakly suppresses
- L5:1952 weakly excites
- L11:2785 weakly suppresses
- L18:2392 weakly excites
- L19:1131 weakly excites
- L7:227 weakly excites word fragments inside split words
- L2:2738 weakly suppresses
- L1:76 weakly excites
- L17:478 weakly suppresses
Feeds later layers
- L26:1055 weakly suppresses
- L27:1721 weakly suppresses
- L27:1560 weakly suppresses
- L26:2912 weakly excites
- L27:918 weakly excites
- L26:370 weakly excites
- L26:1482 weakly excites
- L26:786 weakly excites
- L27:2830 weakly excites
- L27:1558 weakly excites
- L27:2679 weakly suppresses
- L26:1145 weakly suppresses
- L27:2496 weakly suppresses
- L27:1085 weakly suppresses
- L26:2611 weakly excites
- L26:2087 weakly suppresses
- L26:2248 weakly suppresses
- L26:1902 weakly suppresses
- L26:2342 weakly excites
- L26:2095 weakly suppresses
- L27:1400 weakly suppresses
- L27:229 weakly suppresses
- L27:1154 weakly suppresses
- L26:2323 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
- ità
- ilar
- 晚
- ress
- atures
- ýt
- gage
- ymi
- vasion
- adamente
Suppresses
- IService
- 破
- kind
- breaking
- 积蓄
- anga
- <th
- едь
- 服务
- Service