zero before a decimal point
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 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 digit 0 when it sits just before a decimal point, most often in statistics such as p < 0.001. Turning it up made the model output strings of symbols and equals signs, so that test gives little insight.
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.
-
3, 95% CI = 1.110-1.529), local citizens (p < 0.001) and officials/clerks (p = 0.
-
6%) compared to those of white-nosed coati (9.05%), being this difference significant in summer 2012 (P < 0.00001), summer (P =
-
the results were printed to three decimal places. When typing in the new conditions, Lorenz entered the rounded-off numbers (0.506 rather than 0.506127, a difference of less than 0.
-
to directly correlate the LLCCs length to the diameter of the inner tubes and determine the optimal diameter for the high yield growth of carbyne (about 0.71 nm). Our approach of HV high temperature annealing has a great
-
where c = (10/9)2 ≈ 1.2 and x = (9/10)log102 ≈ 0.97. The expected number of powers of 2 without zeroes starting from
-
22.91% of sampled diarrhea patients. The seasonal distribution of norovirus infections was different from non-norovirus patients (p < 0.001), with a half-year peak. Higher proportions of males (
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: the idea that a single individual's actual effort to achieve a certain intent or intent under actual economic conditions, such as actual economic efforts, actual efforts to achieve a certain intent, actual efforts to achieve a certain intent, actual efforts to achieve a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
working contact. The U.S. has operated “interest sections” in places like Cuba and Iran.⏎South Korea set up a liaison office in the North Korean city of Kaesong last year after several months of increased discussions and contacts with the North
-
leading a flight of fighters of the 39th Fighter Interceptor Squadron, 51st Fighter Interceptor Wing, based at K-13, near Suwon, Republic of South Korea. He was flying his usual airplane, North
-
Kentucky, two border states that officially remained in the Union, while the western counties of Virginia rejoined the North. The Confederacy's capital was moved from Montgomery to Richmond, Virginia, in May 1861, and regular presidential and
-
photo of Trump and Kim Jong Un on its front page.⏎Trump and Kim were asked by reporters about the possibility of opening a liaison office during their meeting last month in Hanoi, Vietnam. “It’s actually not a bad idea,” Trump said
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
- L21:351 suppresses
- L20:373 excites
- L20:2980 suppresses
- L23:1291 excites semicolons and list-like citations
- L22:553 weakly suppresses
- L23:209 weakly suppresses
- L22:544 weakly excites
- L6:2561 weakly suppresses
- L3:236 weakly excites
- L21:162 weakly excites
- L18:1132 weakly excites
- L24:127 weakly excites
- L0:2583 weakly suppresses
- L18:335 weakly suppresses
- L21:1829 weakly excites
- L20:1391 weakly excites
- L9:1522 weakly suppresses
- L17:1134 weakly excites
- L20:206 weakly suppresses
- L15:2169 weakly excites
- L18:1187 weakly suppresses
- L18:775 weakly suppresses
- L8:692 weakly excites
- L10:32 weakly suppresses
Feeds later layers
- L27:682 excites
- L27:858 excites
- L26:582 excites
- L26:122 excites
- L27:2125 excites
- L27:646 excites
- L27:390 weakly excites
- L27:2906 weakly excites the letter r in the middle of words
- L27:753 weakly excites
- L26:2614 weakly excites
- L26:606 weakly excites
- L26:712 weakly suppresses
- L27:2214 weakly excites
- L26:2661 weakly excites
- L27:760 weakly excites
- L27:1603 weakly excites
- L27:1424 weakly excites
- L27:3022 weakly excites
- L26:1353 weakly suppresses
- L26:227 weakly excites
- L27:2808 weakly suppresses
- L27:2851 weakly excites
- L27:415 weakly excites
- L26:2455 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
- :
- =r
- =c
- =q
- =z
- =v
- =d
- :
- =
- =m
Suppresses
- .GetEnumerator
- Sous
- economic
- \application
- imizer
- hai
- 着他
- 真情
- her
- activex