the digit 0 in dates and numbers
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. 18 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 inside a year, a measurement or a date range, such as the 0 in 1600 or in 12.0 m. Turning it up gave answers that wandered into unrelated topics, so it does not clearly change the model's style.
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.
-
00–500 BC ) and because of the large number of new foundations in Hellenism (323–30 BC), the Mediterranean world remained urban for centuries, although the majority of the People
-
-e to produce a Sino-Japanese painting style which was dominated by the Kanō school from the sixteenth century. In the Edo period (1600-1868) the Rimpa, Zen and Bunjinga
-
a genre of Japanese art that flourished during the Edo period (1603-1868). It is a form of woodblock printing that
-
at⏎= 56.3m/s + (-9.81m/s2)(4.52s)⏎vf = 12.0 m/s⏎This value is still positive, but smaller. The ball is slowing
-
tons)|⏎|Length:||205 ft (62.48 m)o/a|⏎|Beam:||33 ft (10.06 m)|⏎|Draught:||11.5 ft (
-
is derived from the good old Spring and Autumn period (772BC to 403BC) and the Warring States period right after (403BC to
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 use of a student who is a non-bone, but it can be given to a student. If that's possible, and it is, then a good and appropriate number of them can be given to. Then, we can have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
%) compared to those of white-nosed coati (9.05%), being this difference significant in summer 2012 (P < 0.00001), summer (P =
-
2.91% 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 (p
-
and staying healthy, both physically and mentally.<|endoftext|>Tech moves fast! Stay ahead of the curve with Techopedia!⏎Join nearly 200,000 subscribers who receive actionable tech insights from Techopedia.⏎A vulnerability management system
-
3, 95% CI = 1.110-1.529), local citizens (p < 0.001) and officials/clerks (p = 0.0
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
- L17:213 suppresses
- L2:1795 suppresses
- L16:177 excites zero digit inside dates and codes
- L17:128 excites
- L4:498 suppresses
- L5:1546 suppresses
- L16:827 suppresses
- L10:50 excites
- L10:1052 suppresses
- L16:215 suppresses
- L15:12 excites
- L0:44 weakly suppresses
- L13:33 weakly excites
- L0:2302 weakly excites
- L16:374 weakly excites
- L15:78 weakly excites
- L16:461 weakly excites
- L15:96 weakly suppresses
- L16:724 weakly excites
- L2:1334 weakly suppresses
- L7:1808 weakly excites
- L10:722 weakly suppresses
- L8:20 weakly suppresses
- L17:512 weakly excites
Feeds later layers
- L20:626 excites
- L19:843 excites
- L19:311 suppresses
- L19:246 suppresses
- L19:2478 suppresses
- L19:321 suppresses
- L20:121 excites
- L20:1852 suppresses
- L19:129 suppresses
- L20:1793 suppresses
- L19:974 suppresses
- L20:72 suppresses
- L19:287 suppresses
- L19:197 weakly suppresses
- L19:171 weakly suppresses
- L20:221 weakly suppresses
- L19:2751 weakly suppresses
- L24:1676 weakly excites zeros in large round numbers
- L20:110 weakly suppresses
- L21:280 weakly suppresses
- L20:133 weakly suppresses
- L19:217 weakly suppresses
- L20:217 weakly suppresses
- L20:189 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
- fq
- orable
- 模范
- oration
- حوال
- OfDay
- aurus
- واقع
- ◻
- 年代
Suppresses
- antha
- tures
- ∅
- istics
- реакци
- issant
- двиг
- แต
- /entity
- gest