spot just before a year or century
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. 26 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 spot in a sentence where a year or century number is about to come, such as after 'in the' before '18th century' or 'in' before '2008'. When turned up or down, the model's written output did not show a clear matching theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. 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.
-
at the Uffizi and finally in Palazzo Pitti with the foundation of the Accademia del Cimento in 1657. In the 18th century the Lorraine family kept collecting and the Grand Duke Peter Le
-
replace them.⏎- This low-fertility regime resulted from the fall in the total fertility rate that followed the postwar baby-boom (1946 to 19
-
Believe it or not, the sauce was brought to France by an Italian chef called Catherina de Medici in the 16th century and was originally served as a veloute (with lots of
-
atory to the Benelux Customs Union along with the Netherlands and Luxembourg. This later became the Benelux Economic Union in 1958 and in 2008 the treaty between the three countries was renewed and revised under the
-
area. The great tower or keep is the oldest part of the castle and retains many interesting features.⏎Later remodelling, especially in the 19th century, has added many decorative elements to the castle although retaining much
-
), total government spending at all levels was about 8% of gross domestic product. By 1940, it was approximately 15%. In 2010, it had risen to 36%.
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: what if our world could generate energy both green and green energy? Green energy both green energy and green energy both green energy both green energy both green energy both green energy both green energy both green energy both green energy both green energy both green energy
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
said Lopez at the site, where the temperature reaches 45 C (113 F) at this time of year.⏎The first investigatory work was
-
to the space<|endoftext|>SAFETY & ESCAPE PLAN⏎BEFORE A VIOLENT INCIDENT OCCURS⏎• Teach your children to call 911.⏎• Identify neighbors you can tell about the violence in your life and ask
-
the more secure and complex WPA2, which became available in 2004 and is a common shorthand for the full IEEE 802.11i (or IEEE 802.11i-20
-
15.00 hours, chances are you wouldn't immediately recall. But if you were asked where you were when you heard the news about 9/11 you would probably instantly recall. This is because you have repeated this experience many
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
- L25:492 strongly excites
- L25:2960 strongly suppresses
- L22:217 strongly suppresses
- L25:1452 strongly excites
- L25:1478 strongly suppresses
- L23:617 strongly excites
- L23:1598 strongly excites
- L25:845 strongly suppresses
- L22:844 strongly suppresses
- L23:416 strongly suppresses
- L24:2078 strongly excites
- L24:979 strongly excites
- L21:1882 strongly suppresses
- L22:2405 strongly excites
- L19:1985 strongly excites
- L22:1564 strongly excites
- L24:413 strongly suppresses
- L22:1577 strongly suppresses
- L21:1670 strongly suppresses
- L21:626 strongly suppresses
- L21:2039 excites
- L25:2806 suppresses
- L24:2759 suppresses
- L25:2442 excites
Feeds later layers
- L27:180 strongly excites
- L27:328 strongly suppresses
- L27:902 strongly suppresses
- L27:294 strongly excites
- L27:132 suppresses
- L27:1522 suppresses
- L27:678 excites
- L27:530 excites
- L27:992 excites
- L27:1100 excites
- L27:205 suppresses
- L27:1153 weakly excites
- L27:862 weakly excites
- L27:893 weakly excites
- L27:1157 weakly suppresses
- L27:1422 weakly suppresses
- L27:1087 weakly excites
- L27:448 weakly excites
- L27:1329 weakly excites
- L27:153 weakly suppresses the spot just before a historical number
- L27:352 weakly excites
- L27:435 weakly excites
- L27:572 weakly excites
- L27:1202 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
- يث
- AccessException
- 耻
- ReferentialAction
- embar
- ecome
- adamente
- bble
- 局限
- ثق
Suppresses
- 1
- ₁
- 1
- �
- mười
- ¹
- Eins
- errorHandler
- 🔟
- 十一