'on' before time or track
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. 2 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 lights up on the word after 'on' in phrases about being on schedule or on track, such as 'on time' and 'on track'. Its strongest negative hits are in the middle of longer medical or religious words. Turning it up did not change the writing in a clear way.
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.
-
- Optimisation of their PTI experience⏎- Increases driver’s experience and goes some way to reduce stress and pressure.⏎- Assists the driver to maintain on-time services by reducing passenger boarding dwell time, further helping to reduce safety related incidents.⏎
-
West, but across the continent.⏎We know that statements<|endoftext|>Cataracts are, unfortunately, one of the most common eye problems in pets. They are opacities in the lens of the eye and can affect any age, species, or
-
8.6 years for boys and 7.8 years for girls.⏎Nationally, based on the most recent year of data, the target is not on track to be met.⏎Target data specifications⏎Aboriginal and Torres Strait Islander people
-
and sectors were “on track” with the IEA’s Sustainable Development Scenario, which maps out a pathway to
-
ing on time⏎2. Good germination in mid-August⏎3. A frost protection plan⏎Planting on time⏎Planting your desired crop on time consists of knowing the variety’s “days to maturity.” The days to maturity indicate
-
five ways to prevent summer learning loss and help your student retain information to help them succeed when the new school year begins.⏎1. Create a routine to stay on track.⏎Most students receive summer work from their teachers
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: I want to be a better person, I want to be a better person. So I want to be a better person. So I want to be a better person. So I want to be a better person. So I want to be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are marketed for commercial use. They possess a strength variation because industrial components would just require a strong formula so as to keep the components together. However, home epoxies utilize still and strong formula to keep the stuff together and you can also visit
-
webpage:⏎Protestants should not hesitate to read / study the list contained in the site above. While I was raised thinking that Jesus did not quote the Apocrypha, I was startled several years ago when I found, not direct quotes
-
species into the skin.⏎Common diagnoses treated with iontophoresis include plantar fasciitis, bursitis, lateral and medial epicondylitis (commonly referred to as tennis elbow and golfer's elbow
-
: Resurrection.⏎I came to identify this concept as Maccabean. Book Two of the Books of the Maccabees–found in the Apocrypha and online–chronicles the Jewish war against the Antiochus
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
- L0:1224 weakly suppresses
- L1:1464 weakly excites
- L0:1301 weakly suppresses
- L0:178 weakly suppresses
- L0:1433 weakly excites
- L0:1285 weakly suppresses
- L1:1735 weakly excites
- L0:3033 weakly suppresses
- L0:2421 weakly excites
- L0:2679 weakly excites
- L1:2743 weakly excites
- L1:1960 weakly excites
- L1:1430 weakly excites
- L1:30 weakly excites
- L1:1674 weakly excites
- L1:2759 weakly excites
- L1:1205 weakly suppresses
- L0:2232 weakly suppresses
- L1:2806 weakly excites
- L0:1152 weakly suppresses
- L1:35 weakly excites
- L0:1928 weakly excites
- L1:340 weakly suppresses
- L1:2801 weakly excites
Feeds later layers
- L13:1592 weakly excites
- L8:2551 weakly excites
- L7:2713 weakly excites
- L4:2277 weakly excites
- L10:2135 weakly excites
- L4:1557 weakly suppresses
- L11:2753 weakly excites
- L9:727 weakly excites
- L5:342 weakly suppresses line breaks inside wrapped sentences
- L10:2883 weakly excites
- L12:168 weakly excites
- L9:198 weakly excites
- L15:348 weakly suppresses
- L20:1367 weakly excites
- L13:1639 weakly suppresses
- L8:2397 weakly excites
- L5:2302 weakly suppresses
- L6:2381 weakly excites
- L14:2990 weakly suppresses
- L11:2517 weakly suppresses
- L20:2727 weakly suppresses
- L10:2701 weakly excites
- L9:1284 weakly excites
- L4:1003 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
- /Math
- Disclosure
- bins
- .styleable
- 醪
- nement
- öğretim
- blender
- présence
- RAND
Suppresses
- antar
- *
- anic
- lica
- https
- okin
- .createdAt
- http
- ointed
- pire