the 'lo' fragment in loom and loaves
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 small chunk 'lo' inside words like 'loom', 'loins', 'loaves' and 'loo'. The opposite side responds to the letter 'b' starting words like 'bumpy' and 'bemoan'. Steering did not show 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.
-
the second best with the red cordon, and the third with the green. The thread is girt about their loins, and no ribbon of the Legion of Honour, or Knight of the G
-
wright thought for one moment of the hardship and misery he was about to unleash on a vulnerable section of society, he would never have sat down to invent the power loom.⏎Cartwright (1743-1823) was
-
Museum’s Textile Collection contains over four thousand patent models. The collection includes many examples of carding machines, spinning machines, knitting machines, rope making machines, looms, baskets, carpets, fabrics, and sewing machines. Even the simple clothes
-
doesn’t like to launch things across the room?!⏎Little engineers can use rubber bands to solve a STEM paintbrush engineering challenge!⏎Put together a quick rubber or loom band busy bag to take with you on travels or elsewhere and keep your toddler
-
14, 2004). To the right is a set of rocks referred to as "Rotten Rocks" for their resemblance to rotting loaves of bread. The insides of these rocks appear to have been eroded
-
a surplus of storable food, a system of writing, a more complex social organization, and technological advances such as the plough, potter's wheel, loom, and metallurgy.⏎Cities exist for many reasons, and the diversity of
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: there are some people who make money withlesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylesslylessly
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Israel Sadovnik Socratus. =. P.S. ' They would play a greater<|endoftext|>The big house at the end of the long and bumpy road is rarely quiet. If the house’s occupants are outside, a visitor
-
frequently have the opposite phenomenon, dysphemism, where a relatively neutral word is replaced with a harsher, more offensive one.⏎Such as calling a cemetery a boney
-
taken to complaining about the lack of AM radio. Consumers took to forums to bemoan the omission, a few even saying they wouldn’t buy an i3
-
tois brothers were a mere hundred miles apart; Sévère had set up an independent government on Providencia Island, in the western Caribbean, and Joseph was embarking on a political career of his own in Port-au-Prince, Haiti.
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
- L21:754 strongly excites
- L19:986 strongly excites
- L13:44 suppresses inside web addresses and citations
- L18:2403 excites
- L17:1061 excites
- L19:400 excites
- L10:39 suppresses
- L21:525 suppresses
- L20:541 excites
- L10:428 excites
- L20:54 excites
- L10:155 excites
- L18:178 excites
- L9:17 suppresses
- L19:134 suppresses
- L15:390 excites
- L18:550 suppresses
- L4:342 suppresses
- L12:158 suppresses
- L16:2766 excites
- L18:251 excites
- L23:126 suppresses
- L10:669 suppresses
- L18:233 excites
Feeds later layers
- L27:1916 weakly excites capital P starting a word
- L27:1402 weakly excites
- L27:418 weakly excites
- L27:1786 weakly excites
- L27:1347 weakly excites
- L27:709 weakly excites
- L27:930 weakly excites
- L27:1022 weakly excites
- L27:2148 weakly excites
- L27:2880 weakly excites
- L27:1333 weakly suppresses
- L27:2461 weakly excites
- L27:365 weakly excites
- L27:1984 weakly suppresses
- L27:1583 weakly excites
- L27:2448 weakly excites
- L27:1655 weakly excites
- L27:14 weakly excites
- L27:2080 weakly excites first piece of a split name
- L27:349 weakly suppresses
- L27:461 weakly excites
- L27:1007 weakly suppresses
- L27:2016 weakly excites
- L27:1454 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
- akest
- олод
- �
- enuity
- plets
- emark
- ilateral
- redirected
- uali
- ertas
Suppresses
- """⏎⏎⏎
- lessly
- hood
- 称之为
- gravy
- 市中心
- archy
- …”⏎⏎
- (QObject
- وجب