unclear: weak mid-sentence function 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. 15 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 is weak and hard to read. Its strongest hits are small words such as 'from', 'to', 'but' or 'the' in the middle of a sentence, but the examples have no shared topic. The steering results are mostly repetitive or off-topic, so there is no clear story for 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.
-
009 International Conference on Alzheimer's Disease (ICAD 2009) in Vienna.⏎"We can't do anything about aging or family history, but research continues to show us that there are lifestyle decisions we all can make to keep
-
this article, we will look at five strategies for motivating students and keeping their brains engaged over the summer.⏎Engage in Reading Adventures⏎There is no better time to read than during the dog days of summer. Whether it’s sci-fi, historical
-
it quickly! There are a variety of products dedicated specifically to this purpose:⏎While you’re running or working out in general, your body diverts blood flow away from your digestive system because your muscles need improved blood flow and nutrient accessibility during a workout
-
of the specific query.The Oracle 9i provides the new functionality of With Clause.The with clause is just like temporary tables in oracle.Mainly the with clause is used to execute subqueries in fast way.It is called as sub-query factoring.⏎
-
the beginning.⏎It only consumes constant space.⏎In an imperative language (here Modula-3) the same would be achieved with the following code:⏎TYPE List = REF RECORD next: List; value: INTEGER; END;⏎VAR x := NEW
-
ici published in 1700.⏎Its Afrikaans common name, hangertjie, means 'the little hanger', and is thought to come from the way the sunbirds and sugarbirds appear to 'hang' from the flowers
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 reading about the history of the world, and I have been reading about the history of the world, and I have been reading about the history of the world, and I have been reading about the history of the world, and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Step 1: Create a Shortcut⏎Here you have to create a shortcut of your application.⏎Right-click on desktop click on new then click on the shortcut⏎Then you will see one window.⏎Here you have to paste the full path of your
-
(Applications > Accessories > Terminal) and navigate to the newly mounted drive. All mounted drives should be in /media, so we’ll use the command⏎and then type the first few letters of our difficult-to-type drive, press tab to auto
-
easy as with other programming languages.⏎To create a list of integers, create a new variable and:⏎That’s simple right? And that wasn’t possible with Record Lists.⏎Creating
-
paste it at the following path⏎Now open Run (ctrl+R) and type your command name(Here my command name is “DOSBox“) and hit enter, your application will open. Every time you can do that to open
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
- L12:2653 excites
- L12:1315 excites
- L12:208 excites
- L12:2888 excites
- L14:2863 suppresses
- L11:1151 excites
- L11:641 excites
- L12:271 suppresses
- L10:2955 excites
- L14:2918 excites
- L12:1766 excites
- L14:2122 excites
- L12:3070 excites
- L7:804 weakly suppresses
- L13:1700 weakly suppresses
- L12:2124 weakly excites
- L14:2953 weakly suppresses
- L12:1311 weakly suppresses
- L12:311 weakly suppresses
- L12:822 weakly excites
- L13:1334 weakly suppresses
- L14:1417 weakly excites
- L14:3003 weakly suppresses
- L12:2862 weakly excites
Feeds later layers
- L16:2971 suppresses
- L17:729 suppresses
- L20:643 excites
- L16:1266 suppresses
- L17:1164 excites
- L16:1986 excites
- L16:1995 suppresses
- L16:2788 suppresses
- L16:531 excites
- L16:3016 excites
- L18:2011 suppresses
- L16:2016 excites
- L16:1233 excites
- L17:1033 suppresses
- L19:1517 suppresses
- L16:814 excites
- L17:2405 suppresses
- L16:2837 suppresses
- L16:2010 weakly suppresses
- L16:1311 weakly suppresses
- L17:2110 weakly suppresses
- L16:2605 weakly suppresses
- L16:1882 weakly excites
- L24:1239 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
- icators
- ViewItem
- selectedIndex
- toFloat
- NotFoundException
- strikeouts
- DatePicker
- ketogenic
- visually
- terminated
Suppresses
- ôle
- ariat
- igar
- apologies
- earnest
- owitz
- 接
- 装载
- Bulk
- abler