'for' as in 'for example'
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. 23 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 'for' when it sits in an aside like 'if, for example' or 'for some reason'. Turning it up or down made the model fall into repeated filler words like 'the the the' or 'from from', so the steering results are not very readable.
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.
-
All U.S. states have neonatal screening for CAH. Infants who test positive need to have follow-up testing done to confirm the diagnosis. If, for some reason, the neonatal screening is negative but there is high suspicion for CA
-
⏎How do people determine lengths when they can't use standard measuring tools, such as a tape measure? How do they find the distance across a river if, for example, they don't have a boat? Measurements of very large objects or of
-
US, and Environment Canada - WILDSPACE.⏎- Migration/Movement:Elevational Migrant⏎- Primary Habitat:Montane evergreen forest edge⏎- Foraging Strata:Understory/Canopy⏎- Foraging Behavior:Probe
-
molecules into long, straight rods lets electrons flow freely, approximating the conductivity of traditional materials like silicon or copper. Heeger and his co-Nobelists, for example, discovered that oxidizing the polymer polyacetylene with iodine vap
-
to them.⏎The exhibit tells the story of migrant farmworkers, their work, and the people who serve them, in a number of ways. Photographs, for the most part by the superbly skilled professional photographers who have been acknowledged previously,
-
Inventory Number: AUT 040<|endoftext|>Communities in Kenya have adopted one another’s food preferences and recipes. With the exception of the Coast and Indian communities for whom food preparation is an elaborate process, nearly all other communities garnish their food
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: from of of of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of from of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
get can relies on the amount ingested. My pet has had tiny amounts of onion and garlic and it hasn't caused any issues. He once additionally stole an After-Eight! Mint, once more - this was simply a tiny low quantity of
-
following lesson was created by Vimeo for its Vimeo Video School . It’s used here with permission. Look for a new lesson each week.⏎Want to get started with After Effects but don’t know where to begin? In this lesson Nick Campbell will guide
-
playback the piece you’re working on⏎- Create a pre-comp to easily duplicate certain animations for re-use⏎Here’s another great example video done by Nick in After Effects using precomps!⏎Check back next week to learn the difference between frame
-
and consultations with the general public and initiated public campaigns to encourage people to voluntarily turn in their firearms. The government’s new security initiative “Gran Misión a Toda Vida Venezuela” pledged to continue this work of disarmament, including through 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
- L20:593 excites
- L20:283 excites
- L18:293 suppresses
- L17:303 suppresses
- L22:353 suppresses
- L17:1223 excites
- L18:2153 suppresses
- L4:4 suppresses
- L13:132 excites "of" after a noun
- L20:64 suppresses
- L6:1534 suppresses
- L8:45 suppresses
- L10:132 excites
- L19:1980 excites
- L17:706 excites
- L6:388 excites
- L19:2449 suppresses
- L17:385 suppresses
- L10:272 suppresses
- L12:234 excites
- L17:317 excites 'of' after a noun phrase
- L17:264 suppresses
- L22:554 suppresses
- L7:8 excites
Feeds later layers
- L25:247 excites
- L27:149 suppresses
- L27:474 excites
- L27:108 suppresses
- L27:121 suppresses
- L24:49 suppresses
- L26:693 excites
- L27:663 excites
- L26:71 suppresses
- L26:139 suppresses
- L24:92 suppresses
- L24:116 excites
- L27:262 excites
- L27:168 excites
- L26:631 weakly excites
- L25:802 weakly suppresses
- L24:604 weakly suppresses
- L27:528 weakly excites
- L25:1220 weakly suppresses
- L26:920 weakly suppresses
- L27:244 weakly excites
- L25:306 weakly suppresses
- L27:810 weakly excites
- L26:964 weakly excites "in" after a publication or study word
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
- some
- this
- 一些
- the
- various
- several
- any
- those
- many
- one
Suppresses
- iously
- onian
- 也不敢
- 因为他
- 网首页
- from
- 留给
- 因为
- 说出来
- 眨眼