'Old' as part of a name or phrase
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. 21 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 lights up on the word 'Old' when it starts a name or phrase, like 'Old World' or 'Old Faithful', and on similar words like 'old' and 'heirs'. Its strongest negative cases are about inheritance and succession, like 'heirs' and 'testamentary'. Turning it up did not produce a clear 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.
-
, damask, Cherokee, moss, monthly, musk, and "wild roses."⏎Beloved for centuries for their practical uses as well as their beauty, Old World roses were brought to the New World by European settlers who desired a link with
-
woman, even reaching almost four children per woman in the late 1950s.⏎Developed countries have low fertility, and<|endoftext|>Another edition of The Old Farmer’s Almanac Monthly is here! January is full of fun and
-
duty of eliminating economic inequalities to ensure a reasonable and equitable standard of living, to all citizens of the state.⏎Education, health care, social insurance, housing, old age pensions and medical care have become the responsibility of welfare states. They even bother
-
NASA Remix Challenge. Credit: Cookieater2009 on Flickr.⏎Some people like an adventure, but don’t want to leave their home behind — like old Carl in the movie “Up.” So, if you wanted to go to space
-
Comment on the shape of the distribution. In Problem 28 from Section 3.1, we drew a histogram of the length of eruption of California's Old Faithful geyser and found that the distribution is symmetric. Draw a box
-
cables juts out of the sand next to a public walkway at First Landing State Park. Four small rods stick out horizontally from the sides.⏎Research scientists from Old Dominion University installed the contraption on the dunes of the campground beach last week
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 power of the will to be the only legacy to the children of the will. That's the way of the will. Her, the will, is the only legacy to the children of the will. In the spirit of the will,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
nearest surviving male relative to the French king after the three French Capetian kings, Louis X, Philip V, and Charles IV, all failed to produce male heirs. In order to prevent Edward from claiming the throne,
-
. With little knowledge of the hall prior to the Norman Conquest, documented history starts around the end of the 13 th century, when the hall was passed to the Earl of Chester, a nephew of William the Conqueror.⏎Since
-
of two categories – testamentary or living trusts. A testamentary trust is activated by a provision in the Settlor’s Will at the time
-
, returns from banishment. Because he and the Duchess had been together and had children, his son is believed to be the Duchess’s eldest son and therefore the heir to the throne of Amalfi (or Malfi, as the play
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
- L9:2911 suppresses
- L10:2640 suppresses
- L4:2548 excites
- L15:931 weakly excites
- L6:1894 weakly excites
- L20:871 weakly suppresses
- L9:2035 weakly excites
- L12:12 weakly suppresses
- L11:925 weakly suppresses
- L18:1905 weakly suppresses
- L15:2329 weakly suppresses
- L3:2151 weakly excites
- L3:52 weakly excites
- L20:2423 weakly suppresses
- L18:2729 weakly suppresses
- L10:2762 weakly excites
- L11:2737 weakly excites
- L20:730 weakly suppresses action verbs in how-to steps
- L8:2889 weakly excites
- L14:1684 weakly excites
- L4:1624 weakly excites
- L11:2161 weakly excites
- L10:2328 weakly suppresses
- L11:1035 weakly excites
Feeds later layers
- L23:2059 suppresses
- L24:2075 weakly suppresses
- L24:2602 weakly suppresses
- L24:2256 weakly suppresses
- L25:1564 weakly suppresses word pieces before 'trans' or 'transfer'
- L25:2668 weakly suppresses
- L25:2169 weakly suppresses
- L25:1648 weakly suppresses
- L27:2338 weakly suppresses capital P starting a name
- L25:2851 weakly suppresses
- L23:2060 weakly excites
- L24:2433 weakly excites
- L22:2774 weakly excites
- L25:2843 weakly suppresses
- L25:1348 weakly suppresses
- L22:2342 weakly excites
- L23:923 weakly excites
- L25:2891 weakly excites
- L27:1512 weakly suppresses
- L26:2605 weakly suppresses
- L23:518 weakly suppresses
- L23:1262 weakly suppresses
- L24:2587 weakly excites
- L22:838 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
- OCI
- olygon
- isy
- Movie
- 濉
- aise
- Ai
- Hang
- rij
- ences
Suppresses
- 繼
- 继承
- succession
- 接
- 继
- contin
- sucess
- 传承
- Transmission
- 交接