"the" after certain verbs and prepositions
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. 18 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 "the" in certain fixed spots, such as "like the" or "comes from the". Turning it up made the model repeat phrases about "the problem" or "the time".
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
blizzard shuts down the highway, Phil finds himself trapped in Punxsutawney. He wakes up the next day, only to discover that it's not the next day at all. It's Groundhog Day all over again.⏎via Wikimedia
-
Diagnosis and Treatment Interventions for Racial and Ethnic Minorities<|endoftext|>Lily of the valley, or convallaria, is not a true lily like the Easter or Asiatic lily, but is still toxic to dogs, humans and
-
be rejuvenated in any way.⏎Likewise, damage by harsh scratching cannot be reversed.⏎Scratched tiles will continue to perform<|endoftext|>The word biathlon comes from the Greek word meaning two contests. At the Winter Olympics it means joining two sports together
-
ithridatic War. As a result, Pompey had annexed Syria as a new<|endoftext|>Our Earth is one among many heavenly bodies rotating around a star, the Sun, which gives us light and heat. They all form what we call our
-
and the Israelite people. Curiously, it was the king of Egypt, and not God, or the Israelites themselves, who coined
-
today."⏎It is a question that will resonate with millions living in quarantine today – as people wake up every morning wondering if the day ahead will be any different from the 24 hours they have just endured.⏎But I have a more positive spin
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 problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Missouri in 2015. They are the same facilities as in 2013, except that the Montrose Generating Station has dropped out of the top 10, and the Mississippi Lime Company
-
as heart disease.⏎This study looked only at women diagnosed with breast cancer early – known as stages one or two. It did not include patients with advanced forms of the illness.⏎Experts believe radiotherapy may be far more effective at killing all cancerous
-
– is rooted in a dispute over land claimed by Jews as their biblical birthright and by the Palestinians, who seek self-determination. Despite repeated attempts to end the conflict between the two countries, there is no peace settlement in sight. The issue
-
for you. Thank You.⏎When I first entered the spoken word and writing community in the mid-1990s, one of the biggest names in the Southern California scene was Michele Serros. The Oxnard-born Serros great
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
- L16:792 excites "the" inside a prepositional phrase
- L17:30 suppresses
- L16:419 excites
- L10:1377 suppresses
- L9:662 suppresses
- L12:435 suppresses
- L11:1168 excites
- L16:1320 suppresses
- L4:52 excites
- L9:24 excites "the" before a named organisation or title
- L8:636 suppresses
- L10:120 suppresses
- L17:372 suppresses
- L11:331 excites
- L16:2248 suppresses
- L9:2198 excites
- L13:824 suppresses
- L6:443 suppresses
- L16:2062 excites "other" and contrast words
- L11:875 weakly suppresses
- L10:1644 weakly excites
- L3:344 weakly excites
- L15:1848 weakly suppresses
- L13:1234 weakly suppresses
Feeds later layers
- L21:843 suppresses
- L19:2430 suppresses
- L20:1197 suppresses 'this' pointing at the current document
- L20:1983 suppresses
- L19:1032 suppresses
- L19:2638 suppresses
- L20:714 weakly suppresses
- L19:2537 weakly suppresses
- L19:2958 weakly suppresses
- L23:1817 weakly suppresses
- L19:326 weakly excites
- L21:713 weakly excites
- L24:1416 weakly excites
- L20:47 weakly suppresses
- L19:1900 weakly suppresses
- L19:1159 weakly excites
- L20:337 weakly excites
- L27:533 weakly suppresses
- L22:2298 weakly excites
- L20:2860 weakly excites
- L24:704 weakly excites
- L23:355 weakly suppresses
- L19:1023 weakly suppresses
- L21:639 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
- ason
- InView
- ляр
- &o
- asons
- &a
- courses
- نظم
- ển
- одеж
Suppresses
- ronym
- 事
- 冒
- procedure
- piel
- kh
- 其
- matter
- suck
- ilde