'turn' in 'which in turn'
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. 8 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 'turn' in the phrase 'which in turn', where one thing causes another in a chain. Turning it up pushed the model toward repeating phrases like 'in turn' and 'the temperature of the temperature', so the effect was mostly repetition.
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.
-
reconfiguration region are governed by field-aligned currents, which in their turn generate auroral brightenings near the foot points of the spacecraft. It is
-
, which in turn was divided into the precincts of Beaufort , Hyde , Craven
-
production will be the most affected by climate change by 2030. This would lead to food insecurity, which in turn could lead to social instability, as was already the case in Mozambique in
-
for bacteria which in turn produce methane gas.⏎Methane gas is a greenhouse gas that traps heat much
-
interaction of nature, nurture and the decisions and choices that people make. Biological programming is your genetics. Genes provide the instructions for structuring amino acids which in turn influence the proteins within our body cells thus programming chemical basis of our biology. Our
-
anting trees in clearings and along roads. The communities’ low-impact timber harvesting provides jobs as well as resources for health and education. These opportunities in turn have given the communities a long-term stake in protecting the forest.⏎1 // SITE
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 given to the user a... the user has been given to the user's first time, and it's not related to the time or money. A good value is, I would say that it is... and that it is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
The end result is a vertical wall, or dyke,<|endoftext|>-Why is balance important in an ecosystem?⏎-What is the relationship between living and non-living things in an ecosystem?⏎Then I walked them through
-
ly described by the Monty Python comedy troupe as “taxing foreigners living abroad.” But the fact that
-
’s, a couple of entrepreneurs leased a designer to type a program who post unwanted ads to help you lots and lots of electronic newsgroups. Exactly as Monty Python’s chanting Vikings drowned aside most other conversation into the the latest eatery,
-
is driven by the simple fact that Apache is a hypertext⏎transfer protocol (HTTP) server that runs in the background as a daemon. Because⏎it is a daemon, it must do all the
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
- L7:2495 weakly excites
- L7:2962 weakly suppresses
- L6:2930 weakly suppresses
- L7:2472 weakly excites
- L7:2721 weakly excites
- L7:749 weakly suppresses
- L6:326 weakly suppresses
- L6:757 weakly excites
- L6:2395 weakly suppresses
- L6:1747 weakly suppresses technical word fragments, splicing and isotopes
- L6:2445 weakly suppresses
- L7:1858 weakly excites
- L6:2661 weakly excites
- L4:1929 weakly suppresses
- L2:2703 weakly suppresses
- L7:1251 weakly excites
- L4:2679 weakly excites
- L5:2455 weakly suppresses
- L6:2595 weakly suppresses
- L7:1559 weakly excites
- L4:940 weakly suppresses
- L2:704 weakly suppresses
- L2:2884 weakly excites
- L4:2590 weakly excites
Feeds later layers
- L10:769 excites
- L10:2126 excites
- L16:2338 excites
- L12:2380 excites
- L15:1741 excites
- L14:2295 excites
- L18:1609 excites
- L9:750 excites
- L13:2758 excites
- L12:2564 excites
- L10:908 excites
- L9:2093 excites
- L14:2092 excites
- L17:2861 excites
- L10:2133 excites
- L15:1697 excites
- L9:619 excites
- L11:1973 excites
- L11:925 excites
- L15:931 excites
- L16:1983 excites
- L10:1430 weakly excites
- L11:2786 weakly excites
- L14:275 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
- relations
- juni
- relationships
- relationships
- recursively
- relaciones
- 僭
- ance
- {}.
- angu
Suppresses
- erais
- .eth
- çon
- ắn
- -pres
- deen
- ерь
- 不见
- 粜
- -sided