"at least" 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. 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 fires on the word "least" when it follows "at", as in "at least twice a day" or "at least one". Turning it up did not visibly make the model write about this phrase.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
, at least. The reason? Google and its ever-changing algorithms. Specifically, one of the
-
at least the oft-feared hard landing. The best reply I could give was that it
-
at least 5000 BC, when the area still received substantial rainfall, in
-
to impact memory and contextual learning, which makes it difficult to distinguish danger versus safety. Thus, brain changes in response to early childhood toxic stress could explain, at least
-
clean slate" scenario): (1-2) the two parties in a dialogue, (3) at least one trusted introducer, and (4) at least one trusted witness. Trusted introducers and trusted witnesses allow you to build two open
-
sometimes affect the development of the permanent teeth.⏎Don’t forget that your baby needs more than dental visits to have a healthy mouth. Brush your child’s teeth at least twice a day, avoid sugary drinks or juice, don’t use bottles right
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: they are using a lot of different social media platforms. The new student community at the university has started to use their own social media, and the university has been getting more popular than the old. They are using a lot of different platforms,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, we still want to hear from you.⏎1. University Police will assess the situation for danger of imminent harm.⏎2. A team led by the Dean of Students and the Chief Diversity Officer will review the report and reach out to you to provide
-
pts/pm/research/)⏎The resources below provide activities, resources, and guidance to supporting the mind, body and heart to cultivate positive well-being.⏎Director of Student Services<|endoftext|>The Rape of a Goddess: How Demeter Beat the All-Powerful
-
corridors. Each building has several dust bins from where the housekeeping staff collects the trash. In the same manner waste from canteens, residential quarters, Hostel and guest houses is collected. The collected waste is dumped in big containers (
-
ner will share his personal story of survival and reflect on how the horrors of the Holocaust could have happened in a lecture sponsored by Willamette University’s Jewish Student Union.⏎The free event, which will
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
- L2:111 weakly suppresses
- L2:2345 weakly suppresses
- L3:2794 weakly excites
- L0:2677 weakly excites
- L3:691 weakly excites
- L3:294 weakly excites
- L3:456 weakly suppresses
- L1:2063 weakly excites
- L3:2369 weakly suppresses
- L1:613 weakly excites
- L2:904 weakly excites
- L2:2328 weakly suppresses
- L3:1954 weakly excites
- L2:2021 weakly suppresses
- L3:469 weakly suppresses
- L0:1442 weakly excites
- L0:270 weakly suppresses
- L2:493 weakly suppresses
- L3:2981 weakly suppresses
- L3:867 weakly excites
- L2:49 weakly suppresses
- L2:2122 weakly suppresses
- L0:556 weakly suppresses
- L3:2747 weakly excites
Feeds later layers
- L8:1100 suppresses
- L6:1673 suppresses
- L9:1013 excites
- L5:354 suppresses
- L8:1483 suppresses
- L18:1667 suppresses
- L5:2699 suppresses academic and professional terms, college
- L7:1259 suppresses
- L5:2765 weakly suppresses
- L22:2967 weakly suppresses
- L5:1153 weakly excites 'least' in 'at least'
- L22:1794 weakly excites "shown in" before a figure reference
- L6:1042 weakly suppresses
- L5:1203 weakly excites
- L12:2821 weakly excites
- L6:1037 weakly excites
- L7:2399 weakly excites
- L20:967 weakly excites
- L10:2098 weakly excites
- L5:1686 weakly suppresses the word 'tropical'
- L5:627 weakly excites
- L21:1120 weakly excites
- L5:2100 weakly excites
- L20:506 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
- ian
- çon
- IE
- icha
- ook
- ahun
- hec
- IAN
- 篇章
- breadcrumbs
Suppresses
- rollable
- lạ
- `(
- uates
- физ
- 后排
- `'
- -AA
- bucks
- IFORM