unclear: mixed mid-sentence words
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. 17 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 a wide mix of words in the middle of sentences, such as 'which', 'processors', 'already' and 'complex', with no obvious shared meaning. Turning it up gave repetitive or odd text, so it is hard to say what it does.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
4004 and i4040, Intellec 8 and Intellec 8 mod 80 for the 8-Bit processors.⏎Due to not calling the DEC-PDP-Computer "computer", the Int
-
an older man, seeing each day and the wonders of the natural world as gifts for which we should be grateful. With this video in mind, our teachers show their students
-
cavalry commanded by Lord Cardigan, and to recapture the guns. Lord Lucan and Lord Cardigan saw only the heavy guns in their front—those to which Lord Raglan referred being concealed from their view by the high ground. They,
-
the effects of Word Generation, a middle-school vocabulary intervention, on the learning, maintenance, and consolidation of academic vocabulary for students from English-speaking homes, proficient English speakers from language-minority homes, and limited English-proficiency students. Using individual
-
It’s the first time in recorded history that two strong tropical cyclones have hit Mozambique in the same season.⏎The communities impacted by Idai and Kenneth were already vulnerable to slow-onset climate emergencies, such as droughts, rising sea levels
-
two perpendicular components namely, IRMS cos φ along VRMS and IRMS sin φ perpendicular to VRMS as shown in Figure 4.56.⏎
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) All and $A \rightarrow \{1/2 a_0\+\a_0\ +\b+\2\c_2(2\c_0(2c_0(x=0\+
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
down to horizontal. Sexes and ages are alike.Jackdaws undergo a complete moult from June to September in the western parts of their range, and a month later in the east. Immature birds are duller in plumage.
-
capillary tubes or fixed metering devices.⏎Liquid is then pumped out of the receiver into the liquid line. When the liquid under high pressure reaches the metering device, the cycle starts over.⏎While servicing
-
Write’ function.⏎Do RGB LEDs need PWM?⏎an RGB LED is a combination of three LEDs in just one package: red, green and blue; there are two kinds of RGB LEDs: common cathode and common anode RGB LEDs; you
-
ened and comfortable with initiating and dealing with resistance; it is a mark of true progress.⏎Take a look at what your school is focusing on now and ask these questions: Are we nibbling at the edges or are we considering some real game chang
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
- L2:49 excites
- L6:13 suppresses
- L6:16 suppresses
- L3:115 excites
- L0:1224 excites
- L1:1464 suppresses
- L2:1710 excites
- L0:1301 excites
- L2:75 excites
- L8:7 suppresses
- L6:17 excites
- L3:59 suppresses
- L5:46 suppresses
- L0:178 excites
- L5:110 excites
- L6:12 suppresses
- L6:7 excites
- L8:16 excites
- L8:14 suppresses
- L1:2801 suppresses
- L7:122 suppresses
- L2:126 suppresses
- L1:2683 suppresses
- L1:460 suppresses
Feeds later layers
- L18:624 strongly suppresses
- L18:1488 strongly excites
- L18:823 strongly excites
- L18:1596 strongly suppresses
- L18:962 suppresses
- L18:247 excites
- L20:707 excites
- L18:2064 excites
- L20:677 excites
- L18:755 excites
- L22:245 suppresses
- L19:1773 excites commas in non-English text
- L18:1703 suppresses
- L19:627 suppresses
- L19:1209 excites unclear: commas after dates and clauses
- L18:1764 suppresses
- L18:1956 suppresses
- L21:966 suppresses
- L21:562 suppresses
- L19:788 excites
- L19:767 excites
- L19:2283 excites
- L18:2271 excites
- L20:1159 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
- EFR
- personals
- licts
- licht
- 脲
- 两个人
- categoryId
- ) ⏎
- iphery
- 钋
Suppresses
- a
- A
- ang
- N
- N
- A
- a
- n
- cal
- n