'of' after 'Because'
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. 3 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 "of" in the phrase "Because of" at the start of a sentence. Turned up, the model's text wandered into random topics, with no clear link to the trigger.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
and continue to lower the climate impacts of cotton production.⏎Because of fertilizer’s outsized impact on GHG emissions, Cotton Incorporated has launched a coordinated
-
scholarship above suspicion and his accusers guilty of slander. The damage done his reputation, however, survived him because of historians who failed to report his rehabilitation.⏎Because of his personal qualities as well as his scientific adventures Hell was held in high esteem by
-
the same size. Creating a RAID 5 volume with components of different sizes results in unused disk space.⏎Because of the complexity of parity calculations, volumes with greater than about 20 percent writes
-
"awn," attached to each seed. Beardless varieties are generally preferred for forage, but the bearded varieties have proven resistant to deer in Pennsylvania.⏎Because of the different purposes of the grain, there are many varieties
-
automation. Thanks to its controllable IO (input-output) pins, sensors and other hardware can be read out and controlled very easily. Because of this and its low price, it is also a very popular way to get started
-
results. Governments use it for elections, services and disaster alerts. The global financial system—bank transfers, stock trades—increasingly relies on it.⏎Because of abstraction and the distributed nature of cloud computing, no single data centre holds all 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 been thinking about the importance of my name. I have been thinking about the importance of my name. I have been thinking about the importance of my name. I have been thinking about the importance of my name... and so on,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
summer, create shade with a retractable awning, an umbrella or trees. "Look at the gifts and challenges that nature gives
-
016/index.html<|endoftext|>Glacier Park's cold winters, short growing seasons and high altitudes are less than ideal for reptiles. These animals are ectothermic, meaning they gather body heat from their surroundings. There are only three species
-
world of fun trivia quizzes and quiz games:⏎The Battle of Badr⏎"The Battle of Badr was a key battle in Islam's struggle for independence. Outnumbered greatly, the victory at Badr gave the army great hope and is mentioned
-
Frye, 1956, p. 106 – 119]. However, according to the norms of the Sogdian language, this name could not be transformed from a word vihara for
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
- L2:1880 weakly excites
- L1:2342 weakly excites
- L2:122 weakly excites
- L2:977 weakly excites
- L1:2801 weakly excites
- L2:2979 weakly suppresses
- L1:2154 weakly suppresses
- L0:2645 weakly excites
- L2:549 weakly suppresses
- L2:171 weakly suppresses
- L1:2643 weakly excites
- L1:858 weakly excites
- L1:2249 weakly suppresses
- L1:1588 weakly suppresses
- L2:1873 weakly excites
- L0:1256 weakly excites
- L1:1997 weakly excites
- L0:1873 weakly excites
- L0:1301 weakly suppresses
- L2:1657 weakly excites
- L1:2776 weakly suppresses
- L0:2830 weakly excites
- L0:1224 weakly suppresses
- L2:2565 weakly suppresses
Feeds later layers
- L4:1535 excites
- L9:2052 excites 'because' and 'that' after a reason
- L6:2143 excites
- L18:2841 weakly excites
- L5:914 weakly excites
- L25:2438 weakly suppresses
- L4:1787 weakly excites
- L9:2552 weakly suppresses
- L15:1163 weakly excites
- L23:635 weakly suppresses
- L15:1987 weakly excites
- L4:2117 weakly suppresses
- L9:1465 weakly excites
- L13:3017 weakly excites
- L5:763 weakly excites
- L8:1864 weakly excites
- L16:2206 weakly suppresses
- L10:1344 weakly excites
- L27:1918 weakly suppresses
- L17:3039 weakly excites
- L4:1500 weakly excites
- L16:2043 weakly excites
- L4:712 weakly suppresses
- L25:1004 weakly suppresses
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
- eload
- :=⏎
- latest
- gens
- //=
- изации
- 导致
- overcoming
- 甍
- !!,
Suppresses
- ibrate
- itemprop
- etermine
- ypy
- swore
- ady
- .describe
- dispos
- liber
- 告诉我