"to" after "found" or "said"
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 fires on the word "to" right after words like "found" or "said", as in "was found to be" or "was said to have". Turning it up made the writing loop and repeat, so the steering does not clearly show this pattern.
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.
-
both letters capitalized to distinguish it from cobalt ("Co").⏎The madwoman of Warehouses 12 & 13 named Helena G. Wells believed herself to be THE H.G. Wells the author. Since H.G. Wells - as
-
able to store subcutaneous fat. This was found to be especially the case among males. Men are more sensitive to changes in fat distributions
-
drinking water supplies of 24 major metropolitan areas—from Southern California to Northern New Jersey and from Detroit to Louisville, Ky. Most of the drug sources were found to be in the
-
the Battle of Badr. How many men did the Meccans have fighting?⏎During the heat of the Battle of Badr, the Prophet Muhammad was said to have done what?⏎Thrown sand into the air in the direction of the enemy⏎
-
established the first church in Antioch and is regarded as the founding pope of the Catholic church. Instrumental in the spread of early Christianity, Peter was said to have walked on water, witnessed the “Transfiguration of Jesus” and denied
-
much ‘party’ in such and such a village”. The coming of elections gives fresh opportunities for the crystallization of parties around patrons. Each patron may be said to have a “vote bank” which he can place at the disposal of a provincial
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: the first thing. It's been a while, and the weather is getting warm. So it's been a while. So it's been a while. The first thing is just. The first thing is just. So it's been a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
for osteopenia are lifestyle habits everyone should adopt. If you have strong bones, a healthy lifestyle can help keep them that way. If you already have osteopenia, these same habits can help reduce the risk of osteoporosis.⏎Exercise
-
processes which ultimately result in excess cost, morbidity, readmissions, and mortality.1-4⏎There are two major types of apnea:⏎- Central sleep apnea (CSA) is due to temporary failure in
-
low testosterone, but they are often written off to “stress.” Even clinical depression may be simply due to low testosterone. Men with low testosterone may also have osteopenia or even osteoporosis (thin bones). Testosterone keeps the body skin
-
thinning bones become so fragile that they are prone to fracture easily. A person with osteopenia may benefit from treatment aimed at strengthening bone.⏎How Osteopenia Happens⏎At age 30, bones are as strong and dense
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
- L1:2016 weakly suppresses
- L2:2681 weakly suppresses
- L1:2336 weakly excites
- L0:1301 weakly suppresses
- L0:178 weakly suppresses
- L0:1224 weakly suppresses
- L1:2899 weakly suppresses
- L0:1861 weakly suppresses
- L2:1805 weakly suppresses
- L1:2241 weakly suppresses
- L0:2810 weakly excites
- L1:2011 weakly suppresses
- L1:1333 weakly excites
- L0:852 weakly excites
- L0:1043 weakly excites
- L2:483 weakly suppresses
- L1:2228 weakly suppresses
- L2:70 weakly excites
- L2:2868 weakly suppresses
- L0:2915 weakly excites
- L0:2703 weakly excites
- L0:2539 weakly excites
- L2:199 weakly excites
- L0:1896 weakly excites
Feeds later layers
- L8:997 excites
- L8:2194 excites
- L18:3010 weakly excites 'to' before 'be' after considered
- L15:321 weakly suppresses
- L13:2211 weakly suppresses
- L12:1885 weakly excites
- L5:847 weakly suppresses
- L5:724 weakly excites
- L14:1724 weakly suppresses
- L10:684 weakly suppresses
- L4:730 weakly excites
- L4:2582 weakly suppresses
- L10:109 weakly suppresses
- L8:1096 weakly excites
- L17:582 weakly suppresses
- L19:2067 weakly suppresses
- L6:1974 weakly excites
- L11:3000 weakly excites
- L15:400 weakly excites
- L4:1380 weakly suppresses
- L18:1975 weakly excites
- L10:2732 weakly excites
- L15:1461 weakly suppresses
- L8:2355 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
- dimension
- ptides
- ojis
- Brewing
- /method
- омер
- iedades
- mutations
- [method
- ��
Suppresses
- .Parse
- >manual
- عون
- esis
- ESIS
- 和其他
- 幹
- ML
- Nib
- <?=$