"at" ending a phrase
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. 19 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 "at" in fixed phrases like "not at all", "at a time", "at the same time" and "late at night". When turned up, the model fell into loops of "at the end of the day".
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 water and waste in the system to flow smoothly out to the sewer or septic tank. If there were no vent, the system would drain slowly or not at all, not unlike a water-filled straw when you seal one end with your finger
-
Roughly a third of the people are poor, with 23 percent living “in misery, on one meal a day and not a very nourishing meal at that.” The rich, on the other hand, are very rich indeed. “
-
place cool things for him to explore.⏎Still, kids do need to learn the word “no.” You want to reserve “no” for only a few things at that age–activities that might
-
it difficult to say more than just a few words before pausing to breathe.⏎Weight Training Exercises⏎In weight-training exercises, you generally focus on one muscular group at a time, doing enough repetitions of the exercise to tire the muscles. The Centers
-
is easy to exceed a healthy daily caloric limit when eating fast foods. Eating several smaller meals throughout the day will help you to avoid this type of eating late at night.⏎<|endoftext|>Swept source optical coherence tomography, also known as swept
-
a broad theme that in order to write a profound essay, you need to narrow your topic as much as possible. Otherwise, you will write about everything and nothing at the same time. Think thoroughly what you are asked to do. Does it have
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: at the at a certain at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at at
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an AdobeID at the Adobe website. For more details please see FAQ 1&2. To view
-
or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an AdobeID at the Adobe website. For more details please see FAQ 1&2. To view
-
theological catch-phrase seems to have wide acceptance currently: We are called to be servants. Scriptures seem to invite that identity and it makes sense: Love your neighbor as yourself. Serve others as Jesus served. Let the greatest among you be a servant
-
what exactly happened out there.⏎Thousands of people gathered on the streets of New York, patiently waiting for the arrival of RMS Carpathia…⏎This doesn’t come as a surprise, as the Titanic crash was a disaster of huge proportions back in 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
- L18:895 suppresses
- L4:430 suppresses
- L17:1325 excites
- L15:115 excites
- L15:129 suppresses
- L0:2473 suppresses
- L16:588 excites
- L16:76 excites
- L16:107 suppresses
- L17:166 excites
- L16:260 excites
- L15:78 excites
- L16:567 suppresses
- L1:21 excites
- L7:674 suppresses
- L15:122 suppresses
- L18:102 weakly suppresses
- L16:673 weakly excites
- L15:80 weakly suppresses
- L0:2837 weakly excites
- L8:2038 weakly suppresses
- L18:474 weakly excites
- L16:461 weakly excites
- L16:142 weakly excites
Feeds later layers
- L27:599 strongly suppresses
- L26:835 suppresses
- L23:1501 excites 'at' in idioms like 'at work'
- L22:764 excites
- L26:840 excites
- L24:856 suppresses
- L26:1210 suppresses
- L22:2713 excites
- L24:2924 weakly excites
- L25:501 weakly excites
- L23:2549 weakly excites
- L26:2537 weakly excites
- L21:1872 weakly suppresses
- L21:550 weakly suppresses
- L22:2227 weakly suppresses
- L22:1007 weakly excites
- L22:1493 weakly excites
- L24:1252 weakly suppresses
- L24:2968 weakly excites
- L24:748 weakly excites
- L25:2414 weakly excites
- L27:675 weakly suppresses
- L21:2012 weakly suppresses "to" before an infinitive or noun
- L23:1611 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
- ypical
- tract
- 限り
- Coder
- least
- least
- ienza
- endphp
- emporary
- tracted
Suppresses
- at
- At
- At
- _at
- at
- "At
- atives
- .at
- [at
- -at