'with' after provide or familiarize
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. 4 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 reacts to the word 'with' when it follows verbs like provide, familiarize or 'what could be wrong', where a thing is being given or connected to someone. Turning it up made the writing drift into vague, general phrasing.
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.
-
are very nearly opposite.⏎In his book The Schools Our Children Deserve, Alfie Kohn adds:⏎What could possibly be wrong with results? To answer this question, we first have to recognize that for people to
-
is an issue, and what could be wrong with being an "insider." Some of the papers discuss the issue of objectivity
-
the United States, Canada, Greenland, Russia, Norway, Sweden, and Finland. Researchers hope the study results will provide Native organizations and local governments across the Arctic with information to help make policy decisions. As a major step toward that goal, the
-
is looking at how wind changes or rain could affect the spread of nuclear materials in Japan.⏎"(The center) has been activated to provide the U.S. government with a predictive modeling capability to analyze a variety of potential situations," LaVera said
-
rules include the following:⏎Follow Standardized Symbols - You likely won't get very far trying to read a circuit diagram without familiarizing yourself with the various symbols and their meanings. Different components are usually represented by specific symbols.
-
applied in the clinical setting to help avoid drug calculation and administration errors is boxed throughout the text.⏎- Full-color illustrations, photographs, and drug labels familiarize you with what you'll encounter in the clinical setting.⏎- Current recommendations from The Joint Commission
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 book "A Little Help, a Little Love" (A.L.H.) by Mary R. and her friends. We've been our team, we are, and we're, and we're, and with our team, we are
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
), invented the tactic, and became the first driver to win with it, in the second Daytona 500, in 196
-
018 Chicago Botanic Garden and my.chicagobotanic.org<|endoftext|>Hello. Sign In⏎Please enter a valid city, state or zip code⏎With the Food and Drug Administration (FDA) announcing efforts to reduce salt in the foods
-
47:3; Nahum 3:2; etc.⏎He [God] caused their chariot wheels to swerve, and He made them drive with difficulty; so the Egyptians each said, “Let me flee from Israel, for
-
had a C-section were less likely to do pelvic floor exercises, probably because they didn’t push the baby out with the pelvic floor. But pregnancy itself that affects the pelvic floor, so pelvic floor
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
- L2:863 suppresses
- L3:1218 suppresses
- L3:1968 excites
- L0:2342 suppresses
- L3:187 excites
- L3:2170 excites
- L3:421 suppresses
- L2:2328 weakly suppresses
- L3:272 weakly excites
- L3:186 weakly suppresses
- L3:464 weakly excites
- L3:929 weakly excites
- L2:49 weakly suppresses
- L3:1041 weakly excites
- L3:2277 weakly excites
- L3:691 weakly excites
- L3:189 weakly excites
- L2:237 weakly excites
- L1:2268 weakly excites
- L3:50 weakly excites
- L0:1224 weakly suppresses
- L3:1568 weakly suppresses
- L0:2347 weakly suppresses
- L2:41 weakly suppresses
Feeds later layers
- L9:1856 strongly suppresses "with" after a noun, introducing contents
- L19:2018 suppresses
- L17:2363 suppresses
- L5:910 suppresses
- L9:1340 suppresses
- L16:1860 suppresses
- L6:846 suppresses
- L5:763 suppresses
- L8:2632 suppresses
- L6:2543 suppresses
- L9:2150 suppresses
- L21:1060 suppresses
- L16:2307 suppresses
- L11:1578 suppresses
- L16:1750 suppresses
- L15:2682 suppresses
- L17:303 suppresses
- L10:886 weakly suppresses
- L16:2564 weakly suppresses
- L9:2267 weakly suppresses 'with' after exchange nouns
- L14:1478 weakly suppresses
- L23:3038 weakly excites
- L7:2087 weakly suppresses
- L18:2087 weakly suppresses "with" after a group noun
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
- chemist
- CCA
- �
- ölü
- 很好
- 请输入
- ASIC
- cies
- pla
- 先是
Suppresses
- regards
- regard
- lasses
- tones
- prise
- respect
- Confidence
- Coul
- having
- USTER