word fragments inside non-English words
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. 1 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
This neuron fires on a short piece in the middle of a word from another language, such as 'nos' in a Latvian word or 'als' in a Polish name. Its values are small and mixed, and turning it up did not give a clear pattern 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.
-
products. In the spring of 1934, the Franks r…⏎E-pasta adrese, uz kuru nosūtīt darba saiti:⏎Saite uz darbu:<|endoftext|>
-
is the most obvious similarity between the two religions. In fact a Muslim is defined as anyone who says "Th…⏎E-pasta adrese, uz kuru nosūtī
-
ons of the Prussian King Friedrich Wilhelm I, who finally expelled them from the district. The [[Amsterdam Mennonite Library (Bibliotheek en Archief van de Vereenigde Doopsgezinde Gemeente te
-
inct in the Wild until 2008, but since a series of reintroductions, have developed free ranging populations in the wild. All Przewalski’s horses alive today are descendent of 14 individuals from a captive
-
established to guide players, and the people that coach them, through a level-appropriate system of basketball development. Using scientific guiding principles developed by coach educators Istvan Balyi and Richard Way, and found in their book Long-Term Athlete
-
na in the wilderness (John 6:49-51).⏎Dragons still
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: this is a new question in the field of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ментарии при наличии в них нецензурной брани, оскорблений или спама<|endoftext|>Conflict and Solidarity in Mudejar
-
ЛЬНАЯ РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎КО
-
simple problem, one that a fifth grader may find in a math book. What about the problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 13.09
-
under the czar’s vicegerent in Warsaw. Rzewuski edited the government newspaper Dziennik warszawski.⏎WORKSDziela, vols. 1–6. Warsaw, 1877–8
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
- L0:178 suppresses
- L0:1224 suppresses
- L0:1301 suppresses
- L0:1096 weakly suppresses
- L0:894 weakly excites
- L0:2942 weakly excites
- L0:1 weakly suppresses
- L0:91 weakly suppresses
- L0:794 weakly excites
- L0:3070 weakly suppresses
- L0:104 weakly suppresses
- L0:2609 weakly excites
- L0:2569 weakly excites
- L0:25 weakly excites
- L0:386 weakly suppresses
- L0:2415 weakly suppresses
- L0:188 weakly excites
- L0:2642 weakly excites
- L0:356 weakly suppresses
- L0:540 weakly suppresses
- L0:1315 weakly excites
- L0:63 weakly excites
- L0:1539 weakly excites
- L0:1936 weakly suppresses
Feeds later layers
- L2:87 excites
- L3:809 suppresses
- L5:763 suppresses
- L20:1245 excites
- L17:385 weakly suppresses
- L26:524 weakly excites
- L22:948 weakly excites
- L6:2567 weakly excites "by" after a passive verb
- L10:2023 weakly suppresses
- L18:1135 weakly suppresses
- L14:1353 weakly suppresses
- L17:303 weakly suppresses
- L13:1497 weakly excites
- L7:132 weakly suppresses
- L24:2014 weakly excites
- L19:1278 weakly excites
- L7:572 weakly suppresses
- L3:232 weakly suppresses 'on' after a noun like 'impact on'
- L15:2557 weakly excites
- L20:2014 weakly suppresses
- L16:1727 weakly excites
- L5:1861 weakly excites
- L22:2753 weakly suppresses
- L23:1586 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
- groom
- 不多
- phấn
- poly
- crawler
- unger
- atology
- ivery
- тся
- vation
Suppresses
- uit
- 日前
- recent
- eking
- est
- 日正式
- host
- behalf
- host
- heck