"is" after "What" in questions
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 the word "is" when it comes right after "What" in a question, such as "What is pasteurization?". Turning it up didn't produce a clear change in the written text.
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.
-
frequently asked questions.⏎What is curling?⏎Curling is a bit like a low-speed, two-dimensional
-
fiberglass composite, plastic or aluminum. In addition to crushable brackets and a bar, bumpers sometimes incorporate polypropylene foam or formed thermoplastic.⏎What is a bumper absorber?⏎The bumper absorber is located behind the decorative plastic bumper
-
our economy. Visit America Recycles website at www.americarecyclesday.org to learn more about this subject.⏎Q: What is recycling's greatest economic benefit?⏎A: In a broad sense, recycling is part
-
, and functionality. In this article, we will delve into the main differences between microcontrollers and FPGAs, as well as their advantages and disadvantages.⏎What is a Microcontroller?⏎A microcontroller is a small computer on a single integrated circuit
-
What is the highest overlook on the rim of the Grand Canyon?⏎Point Imperial, 8
-
What is an SOFC?" sidebar below for more information), the PNNL team wanted
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 have been thinking about this one book, and I have been thinking about this one book. I have been thinking about this one book. I have been thinking about this one book. I have been thinking about this one book. I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ovsky Poland⏎Vladimir-Volynsk, Poland… Summer 1943 – Galina and Yuri Miask
-
Litovsk geve the people peace but the price was high and added to discontent causing Civil
-
emyon Alekseyevich Fedorets, commanding the 913th IAP (Istrebitel’nyy Aviatsionnyy Polk, Fighter Aviation Regiment), 32nd IAD (I
-
foods or supplements.<|endoftext|>Bagdannen, a former Mennonite congregation near Tilsit, East Prussia (now Sovetsk, Kalingrad Oblast, Russia), also called "the scattered church of
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
- L1:1412 excites
- L1:1651 suppresses 'of' after 'full'
- L0:1224 suppresses
- L0:178 suppresses
- L0:1301 suppresses
- L2:1770 suppresses
- L1:1029 weakly excites
- L0:792 weakly excites
- L2:41 weakly suppresses
- L2:34 weakly excites
- L2:2328 weakly suppresses
- L2:1238 weakly excites
- L2:483 weakly suppresses
- L0:759 weakly suppresses
- L0:2580 weakly excites
- L0:552 weakly excites
- L0:2457 weakly excites
- L2:20 weakly suppresses
- L0:121 weakly excites
- L2:127 weakly suppresses
- L2:5 weakly excites
- L2:103 weakly suppresses
- L2:2746 weakly excites
- L2:1899 weakly excites
Feeds later layers
- L18:372 suppresses
- L4:1623 suppresses
- L4:800 suppresses
- L24:2431 suppresses
- L4:1473 suppresses
- L9:552 weakly suppresses
- L6:293 weakly suppresses
- L5:1127 weakly excites
- L18:2560 weakly suppresses
- L7:233 weakly suppresses
- L5:90 weakly suppresses
- L5:983 weakly suppresses
- L4:1633 weakly excites
- L6:2716 weakly excites
- L4:1375 weakly suppresses
- L9:815 weakly suppresses
- L4:609 weakly excites
- L12:2921 weakly suppresses
- L8:1266 weakly suppresses
- L10:486 weakly suppresses
- L12:102 weakly excites
- L26:1535 weakly suppresses
- L7:2245 weakly excites
- L12:1581 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
- itivity
- subcategory
- Apprec
- erable
- McC
- mong
- 优于
- til
- ulary
- ingly
Suppresses
- 委书记
- .spi
- òn
- inflamm
- _si
- _();⏎
- ysts
- astr
- äs
- (((