'n' inside word fragments
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. 27 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 responds to the letter 'n' when it appears in the middle of a word, often as the start of a fragment such as 'nkalaweeni', 'nife', 'nose' or 'narrative'. Its negative side is weak and mostly about odd letter breaks. Turning it up gave repetitive text with no clear link to the letter.
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.
-
led out of seclusion and back into the community. Before festivities can begin, the head teacher (kahyuudi or kayudi) commissions a carver (nkalaweeni or mvumbwa) to create a series of masks.
-
tumor from various angles for a short period of time. Stereotactic radiosurgery (SRS) can be performed by several other machines such as X-Knife, CyberKnife and Clinac. It is used to treat brain tumors and
-
al (head)||as in per “capita”, captain, capital|⏎|akshi (eye)||occipital, occular|⏎|nasika||nose, nostril, nasal|⏎|danta (tooth)||dental
-
regions of the country.⏎It is difficult to present any exact etymological meaning of the name ‘Nandinagari’. The first part of the term ‘nandi’ is rather ambiguous in the present context. It may mean ‘sac
-
havaktra Dakini⏎- The Buddhist protector deity Penden Lhamo⏎- The Goddess of the White Umbrella (the Buddhist deity Ushnisha-sitatapatra)⏎- The Great mystic Virupa⏎- Thunder
-
timber rattlesnake, the massasauga (found only in the Rochester/Syracuse area) and the copperhead. “Timber rattlesnakes are generally found in deciduous forests in rugged terrain.⏎What States Have The
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 countryessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessessess
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
research labs and other things of interest⏎Posted: June 27, 2009⏎Nanoscale holes provide speed boost for diagnostic tests⏎(Nanowerk News) Microfluidic devices, essentially miniaturized chemical laboratories
-
the betterment of their students. Once the lesson plans and deadlines are decided, teachers get a roadmap of what to do throughout the year. This helps eliminate vagueness in action
-
volume of knowledge is doubling at least every 8 years. This rate of expansion overwhelms our capacity to stay abreast of emerging knowledge, even in the ever-narrowing micro-specialization of medical practice. New careers are created while other fields
-
’s Head⏎Top female stylists share their favorite men’s looks.⏎10 Kung Fu Movies Every Man Should See⏎From the absolute classics to the<|endoftext|>(Nanowerk News
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
- L16:55 strongly excites
- L19:595 strongly excites
- L2:1923 strongly suppresses
- L18:282 strongly excites
- L17:391 strongly excites
- L1:27 strongly excites
- L22:2299 excites
- L21:567 suppresses the word 'in' after a verb or noun
- L0:210 excites
- L10:1745 excites
- L5:499 excites
- L3:410 suppresses
- L21:2959 excites
- L0:1484 suppresses
- L14:392 excites
- L2:7 suppresses
- L0:24 suppresses
- L19:231 excites
- L16:118 excites
- L1:1110 excites
- L1:99 excites
- L18:127 excites
- L4:2261 suppresses
- L4:221 excites
Feeds later layers
None: this is the last layer.
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
- 徒
- 包袱
- kp
- aghetti
- ��
- 不死
- 屉
- 不如
- .Hash
- 卯
Suppresses
- ess
- ow
- et
- etAddress
- ums
- ews
- ую
- ode
- о
- ot