the last zero in large numbers
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. 3 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 lights up on a zero inside a big number, like the 0 in 1500, 2000 or 800. When it was turned up, the model's replies drifted into rambling text with no clear link to numbers.
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.
-
and brain.⏎What Vitamins Should heavy drinkers take?⏎- Include 250mg Vitamin C, 150mg magnesium, 1500mg calcium and 500 mg niacin from dietary sources each day.
-
] stamp, this stamp is supposed to certify the produce is 90% organic non GMO. So why no seeds?⏎viewed 15008 times⏎People have been modifying plants for thousands of years. This is called
-
War II.⏎Here are some Fw 190’s specs at a glance:⏎- Maximum speed-408 mph⏎- Range-500 miles⏎- Service
-
1 / √2⏎E) 2 √2⏎What is the centripetal force acting on a car of mass 2000 Kg turning around a circular bend of radius 50 meters at the constant speed
-
so they are very clean. You can also disinfect them with a mild bleach solution.⏎- yoghurt containers – the small 8 ounce (250ml) single serving sized ones are
-
DPI laser mice are very popular, but some high-end models are capable of 1000, 1600 or even 2000 DPI tracking speeds.⏎The Personal System/2 or PS/2 was the designation
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 am not a fan with this. I have been on the beach of this thing. So, the first thing is with my head. I am not a person. I'm a person. Okay, this is about me. I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
There are many calculators out there which can help you determine your energy expenditure. One that seems<|endoftext|>Italian scientists have found another image on the back of the Shroud of Turin, reported The Guardian yesterday. The scientists from the University of Pad
-
the result of an intense source of energy as the body passed through the cloth, perhaps at the moment of resurrection,” reported The Guardian.⏎Believed to be shroud of Jesus⏎The man, whose image appears on the shroud, seems to
-
ies, even if abandoned, in ruined condition or transformed for other use, also served as symbolic mezuzahs to mark towns, villages, cities, and even countries where Jews once
-
Stevenson House was the centre of his world and his mind was nourished by ceaseless reading as well as the stories told by Cummie of ghosts, ghouls, Scottish history, and the Bible. He also travelled with his father around
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
- L0:1301 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L1:2659 weakly excites
- L2:1795 weakly suppresses
- L2:79 weakly excites
- L2:1377 weakly suppresses
- L0:2825 weakly excites
- L2:1657 weakly excites
- L2:237 weakly excites
- L1:2544 weakly suppresses
- L1:2411 weakly excites
- L0:2352 weakly excites
- L2:2716 weakly suppresses
- L1:469 weakly excites
- L2:462 weakly excites
- L2:366 weakly suppresses
- L0:2721 weakly excites
- L0:252 weakly excites
- L1:2057 weakly excites
- L0:1646 weakly suppresses
- L2:49 weakly suppresses
- L0:16 weakly excites
- L2:1268 weakly excites 'ur' and 'ud' after a capital H
Feeds later layers
- L4:2836 weakly excites
- L16:1962 weakly suppresses
- L7:1218 weakly suppresses
- L25:2025 weakly excites
- L6:2516 weakly suppresses
- L8:3000 weakly excites
- L14:2085 weakly excites
- L11:1552 weakly excites
- L18:2666 weakly suppresses
- L10:672 weakly suppresses
- L9:750 weakly excites
- L18:2908 weakly excites
- L23:324 weakly suppresses
- L7:2017 weakly excites
- L15:2715 weakly excites
- L10:95 weakly suppresses
- L27:961 weakly suppresses
- L11:1674 weakly excites
- L20:631 weakly suppresses
- L10:476 weakly suppresses
- L7:514 weakly suppresses 'always' or 'often' after 'is' or 'there is'
- L4:2550 weakly suppresses
- L7:1169 weakly excites
- L7:2232 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
- 脲
- Raqqa
- レーション
- IRM
- plies
- ."</
- leitung
- berman
- sWith
- ta
Suppresses
- Voor
- ius
- 凡
- Stat
- Accept
- prom
- ény
- [m
- nost
- eway