"past" or "passed" after motion verbs
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 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 words like "past", "passes" and "by" when something moves or time goes by, as in "slip past your defenses" or "months went by". Turning it up made the text wander into unrelated stories and memories.
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 some point, you are going to get sick. Despite taking such precautions as frequent hand washing or carrying around hand sanitizer, sometimes a germ sneaks its way past your defenses and lodges its way in your system. When that happens, you
-
UR: I HAVE the honor to inform you that Florida and the Lucayan Islands are situate on one side of the Bahama (old) Channel, which passes between Havanna (Cuba) and Florida. But nearer the […]⏎Con
-
See for Yourself⏎If you just want to glimpse the buttes as you whiz past in your car, the easiest thing to do is to simply take a drive along
-
Just as human germs vary, your computer may become infected with something minor - like a trojan - or major - like a worm. When something does slip past your defenses and infects your system, there are steps you can take to get
-
of an Indian chieftain first appeared on the supporters’ flag of the KAA Gent football club in 1924⏎Nevertheless, many years passed between Buffalo Bill’s visit to Belgium in 1906 and the use
-
, and Gundaphorus gave him a large quantity of money for building the palace, and continued to send him large quantities of money and provisions as the months went by. But Thomas took all the money and provisions he received from Gundaphorus and
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: a bit. The problem is to find a balance of a static system with some sort of aspect. The problem is to find a very basic system with a single population. The population is initially filled with with the first reservoir of a chemical system
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
something new together — cheer for the good guys and boo the bad ones in the books you read. You also get a peek into how your child sees the world through the comments she makes on the plot, the characters, and the setting. “
-
exploring the cosmos through a variety of telescopes that they acquire, maintain, and improve on their own. Some of these amateur astronomers specialize in capturing what is seen through their telescopes in images and are astrophotographers.⏎What happens when the
-
proofs, not through the formulaic method that seems to be so popular right now. Mr. Sheen bases his exploration of geometry on the tenet that experience through drawing is of great value to a student, and I couldn’t agree more.
-
Becoming good at math begins long before a child enters school. Even babies can understand simple math, such as noticing patterns. Toddlers love learning math through puzzles and games, such as comparing sizes and shapes. These concepts help them make
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
- L3:1963 suppresses
- L3:1545 weakly suppresses
- L2:2671 weakly suppresses
- L3:50 weakly excites
- L3:2111 weakly excites
- L3:1502 weakly excites
- L2:1657 weakly excites
- L3:2071 weakly excites
- L3:2989 weakly suppresses
- L2:70 weakly excites
- L2:2096 weakly excites
- L3:2162 weakly suppresses
- L0:947 weakly excites
- L1:2944 weakly excites
- L1:2131 weakly excites
- L0:2277 weakly excites
- L3:2615 weakly suppresses
- L2:2963 weakly suppresses
- L3:186 weakly suppresses
- L1:491 weakly excites
- L3:2700 weakly suppresses
- L2:3000 weakly excites
- L3:1860 weakly excites
- L0:768 weakly suppresses
Feeds later layers
- L7:2586 excites
- L8:1484 excites
- L19:2420 excites
- L5:2602 excites
- L25:2169 weakly excites
- L20:1122 weakly excites
- L5:1705 weakly suppresses
- L6:2014 weakly excites
- L7:2374 weakly suppresses
- L19:685 weakly suppresses
- L25:1646 weakly excites
- L17:1139 weakly suppresses
- L15:1066 weakly suppresses
- L11:439 weakly suppresses
- L6:2958 weakly excites the colon after Note at the start of a note
- L5:1050 weakly suppresses
- L15:1358 weakly excites 'process' after 'is a' definitions
- L10:1553 weakly suppresses
- L21:2895 weakly excites
- L7:2255 weakly excites
- L5:2567 weakly suppresses
- L18:1805 weakly suppresses
- L11:2958 weakly excites
- L19:683 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
- yü
- passing
- passes
- Boeh
- oshi
- 多年的
- Woche
- registr
- entity
- MANY
Suppresses
- zig
- .RequestBody
- ièrement
- azaar
- @ResponseBody
- uncated
- 在线咨询
- 聚焦
- -sided
- responseObject