spot where a number or 'time' comes next
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. 21 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 fires when the next word is likely to be a number, such as an hour, a day count, or a duration, like the spot before '4:30 p.m.' or '13.7 days'. It goes strongly negative in phrases like 'without wasting time', where the next word is likely 'time'. Turning it down made the model repeat the word 'time' over and over.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
By wearing glasses that block blue light allows the flow to start (2). The average time for melatonin to flow (if the person is in darkness) is 11.4 hours according to a recent study
-
.⏎Visit our Ed Clark Museum of Missouri Geology on Geologic Map Day to learn more about geologic maps. Open weekdays from 8 a.m. to 4:30 p.m., we are located at 111 Fair
-
high-voltage transients and high operating temperatures, the latter resulting in thermal runaway and over-current conditions. Protection circuitry is straightforward and easy to implement but adds to the bill-of-materials (BoM) and takes up more space
-
unter house telescope TESS, the all-sky search is on for presumably liveable planets near our photo voltaic system. TESS — orbiting Earth every 13.7 days — and ground-based telescopes are poised to search out
-
many children stay up late and sleep in, which can make it difficult to adjust to the early mornings and structured schedules
-
they first arrive. Some come to the centre hungry (it provides meals) or speaking no English. Others arrive with behaviour problems. Parents’ work schedules are often so inflexible that Federal Hill must cover basic logistics beyond school pick-up and drop-off
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: time to save the time. I have a lot of time, but I have to spend time with the time. Time is the time I have. I have a lot of time, but I have to spend time with the time. The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
A? It’s OK. You’re going to P-R-A-C-T-I-C-E and get very comfortable with the test format so you don’t have to waste time reading instructions during the test, right? Also, this acquired familiarity is going
-
this method will help you achieve a bouncing ball animation in probably 5 minutes or less.⏎So without wasting much time, let’s get into it!⏎Here are the steps:⏎- Delete
-
this article. Of course, they can.⏎You’ll surely learn something from these moral stories at the end.⏎Read some other English moral stories.⏎Hence without wasting your valuable time, let’s start writing some of the best moral stories in English
-
70°C. [Read more…]⏎Nowadays, attend system in schools and colleges is generally based on paper. So sometimes this process causes errors and also takes more time. So this project uses RFID technology to make a note of every student
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
- L15:2781 strongly suppresses
- L18:3067 excites
- L19:2792 suppresses
- L17:3016 excites
- L10:1653 suppresses
- L18:2776 excites
- L19:2896 excites
- L15:833 suppresses
- L20:1150 excites
- L15:1823 excites
- L20:1335 suppresses
- L18:1546 suppresses
- L20:1199 suppresses
- L8:1257 suppresses
- L10:1566 excites
- L17:2750 suppresses
- L18:1750 suppresses the blank slot before a year
- L11:1507 excites
- L18:569 excites
- L11:951 excites
- L19:1233 excites
- L19:2989 suppresses
- L19:2706 weakly excites "per" in rate phrases
- L20:2382 weakly suppresses
Feeds later layers
- L27:836 strongly suppresses
- L25:565 strongly suppresses
- L24:2325 strongly suppresses
- L22:1074 strongly suppresses
- L23:1549 strongly suppresses
- L26:2065 suppresses
- L22:2932 suppresses
- L26:2567 suppresses
- L22:2445 suppresses
- L23:2577 suppresses
- L24:1929 suppresses
- L22:2575 suppresses
- L23:2882 suppresses
- L27:934 suppresses
- L22:1563 suppresses
- L24:906 suppresses
- L26:2126 suppresses
- L23:1490 suppresses
- L23:2178 suppresses
- L22:2729 suppresses
- L22:2601 suppresses
- L24:1778 suppresses 'a' before a time span
- L23:2359 suppresses
- L24:1518 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
- ildo
- -alist
- headquartered
- oidal
- حزب
- ervlet
- Saddam
- addError
- uitar
- campuses
Suppresses
- time
- 时间
- 时间和
- time
- (time
- 的时间
- Time
- Time
- 時間
- 时间