single letter starting a split word
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. 26 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
It fires on a lone letter, such as 'o' or 'f', that begins a word which is split into pieces, like 'o' in 'opossum' or 'f' in 'fidgeting'. Steering did not produce a clear theme.
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.
-
for your yard, but it’s diet of fruits and vegetables can also make it a pest that destroys your plants and any vegetable garden. The opossum keeps to itself, traveling alone, and is not a pack animal.
-
, terminal and lateral; bracts and bracteoles absent; rays numerous, upwards-spreading, unequal. Calyx teeth obsolete. Petals yellow, obovate, mid rib conspicuous, apex with narrowly inflexed lobule
-
rets with olive oil, garlic, and parmesan cheese and roast in the oven for a crispy and flavorful side dish.⏎- Broccoli and mushroom omelette: Saute broccoli florets and sliced mushrooms and fold them into a
-
from a medical condition that needs treatment.⏎Common symptoms of anxiety would be increased heart rate, hyperventilation, sweating, nervousness, weakness, fatigue, fidgeting, inability to concentrate, nausea or other stomach related issues, and difficulty
-
, parabolas, ellipses, hyperbolas, cubic curves like elliptic curves and quartic curves like lemniscates, and Cassini ovals. A point of the
-
a dog’s leg and break the ankle of a human, and the poison causes gangrene if untreated. They live along the banks of rivers and oasises.⏎They ambush creatures, hiding in the sand and snapping when a foot
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: people shouldn't be allowed forthtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingtainingt
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
exploitation and human trafficking in Thailand.⏎Human trafficking has become a well-known topic and a well-loved cause during the past several years, but it still remains one of the fastest
-
tics. The effect may only last for a few months, and repeated treatments might result in tolerance, rendering the drug ineffective after several applications.⏎Treatment of Attention Deficit Hyperactive Disorder (ADHD)⏎ADHD is not uncommon in children with
-
“Interdisciplinary System for Interactive Scientific and Scientific-Technical Information”.<|endoftext|>An team of researchers heading to the Antarctic later this year may go in search of a long-lost piece of exploration history. A group of international climate scientists will travel into the
-
each chakra represents? Here is a simple diagram and list for you to follow. | WIKIPEDIA⏎THE CHAKRAS ARE AS FOLLOWS⏎- ROOT/BASE (MULADHARA) CHAKRA
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
- L23:2981 suppresses
- L22:369 excites
- L16:177 excites zero digit inside dates and codes
- L18:437 excites
- L16:67 excites
- L17:213 suppresses
- L10:90 suppresses
- L5:1546 suppresses
- L2:1622 excites
- L18:112 suppresses 'over' meaning time or duration
- L0:44 suppresses
- L1:15 suppresses
- L21:381 suppresses zeros inside long number codes
- L2:1795 weakly suppresses
- L18:328 weakly suppresses
- L17:378 weakly excites
- L2:1787 weakly suppresses
- L5:926 weakly excites
- L7:2095 weakly suppresses
- L17:1046 weakly suppresses
- L20:211 weakly excites
- L10:1052 weakly suppresses
- L7:287 weakly suppresses
- L19:702 weakly excites
Feeds later layers
- L27:686 weakly suppresses
- L27:903 weakly suppresses
- L27:346 weakly suppresses
- L27:2684 weakly suppresses
- L27:210 weakly suppresses
- L27:455 weakly excites
- L27:2875 weakly excites
- L27:420 weakly suppresses
- L27:1435 weakly suppresses
- L27:2857 weakly suppresses
- L27:918 weakly suppresses
- L27:1427 weakly suppresses
- L27:2633 weakly suppresses
- L27:1386 weakly suppresses
- L27:654 weakly suppresses
- L27:2118 weakly excites
- L27:1068 weakly suppresses
- L27:488 weakly excites
- L27:493 weakly suppresses
- L27:652 weakly suppresses
- L27:2374 weakly suppresses
- L27:929 weakly suppresses
- L27:1940 weakly excites
- L27:1825 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
- 郦
- opol
- oug
- nom
- lugar
- EFR
- books
- ynchronized
- ENTE
- ogl
Suppresses
- ichi
- aining
- nds
- t
- ungen
- te
- icho
- tool
- 局面
- aight