"as" completing an "as ... as" comparison
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. 6 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 "as" in the second half of comparisons like "not as harshly as" or "as well as". It also goes strongly negative on words like "least" and "single" that appear in negated phrases such as "not the least suspicion". Steering did not produce text that clearly follows 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.
-
on the 'wrong' side of the scale. But, in results, the group who had washed their hands was less likely to judge the actions as harshly as the group who had not.⏎In another experiment, a group was asked to read
-
of the body, including the brain. As a result, some people may notice that it takes longer to learn new things, they don't remember information as well as they did, or
-
a proper breakfast you won’t perform as well mentally – you won’t be as attentive, your memory won’t be as good and you won’t concentrate as well as you could. Children who eat breakfast on a regular basis tend to perform better acad
-
00.00. Within four months, these men<|endoftext|>Our discourse by now may have proved to the reader that the⏎British Empire is not as mythical as the propaganda would have us believe.⏎We quote from THE NATIONAL MESSAGE, of London
-
may not have the knowledge to present their content with the most efficiency.⏎Bottom line, if you’re not connected to the Internet, you’re not learning as effectively as you could be. Face-to-face interaction is important for students. No one is
-
(a new kind of machine-based conflict).⏎The background to Britain’s entry into the war, he states, was not one of weakness, isolation and austerity, as is often argued, but of an abundance of key resources. During the conflict imports
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: in the house of my possession of a house of ground house of a building of a wall of a house of wall has been occupied. of a building of a structure of a plot. of land of ground. of wall of the structure of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
were supposed to be yearning for the time when the Church should be “purified”.⏎For two hundred years, while the world never had the least suspicion of their existence, the brotherhood transmitted by these means the wisdom of “
-
Physics Mach 1, guided with PHD Guide⏎HA 6 x 15 minutes<|endoftext|>Have you ever sat down to begin writing and not have a single idea about what to write? We are learning to generate ideas in different ways so
-
is one of the express prophecies of the Quran, which proved literally true, afterwards, whereas when this prophecy was made there seemed not to be the remotest chance that the helpless and powerless man who had come out<|endoftext|>- Study Skills.⏎
-
oring and people could find it difficult to breathe," he explained.⏎Therefore, if you have a persistent cold, it is not worth ignoring it.<|endoftext|>Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox
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
- L5:264 suppresses
- L2:899 weakly suppresses
- L1:1661 weakly suppresses
- L0:291 weakly excites
- L3:964 weakly excites
- L3:2653 weakly suppresses
- L4:12 weakly suppresses
- L2:1375 weakly excites
- L4:19 weakly suppresses
- L2:2242 weakly suppresses
- L1:2333 weakly suppresses
- L4:101 weakly excites
- L1:2680 weakly suppresses
- L4:1625 weakly suppresses
- L3:2039 weakly excites
- L3:2211 weakly excites
- L0:178 weakly suppresses
- L1:1674 weakly excites
- L5:57 weakly suppresses
- L3:1003 weakly excites
- L4:1338 weakly suppresses
- L5:198 weakly excites unclear: weak "and" in lists
- L2:681 weakly suppresses
- L3:1812 weakly excites
Feeds later layers
- L7:2272 strongly suppresses
- L7:1103 excites 'as' in 'not as ... as' comparisons
- L7:2514 excites
- L13:1722 excites
- L7:1617 suppresses
- L7:213 excites
- L13:1852 suppresses
- L7:2141 suppresses
- L8:1319 weakly excites
- L18:1478 weakly excites
- L24:3054 weakly suppresses
- L8:3009 weakly suppresses
- L17:2783 weakly excites
- L22:1744 weakly suppresses
- L21:3046 weakly suppresses 'at' in 'not at all' and 'at all'
- L7:1401 weakly excites
- L15:359 weakly excites
- L11:2067 weakly suppresses
- L15:416 weakly suppresses
- L13:1317 weakly suppresses
- L15:1683 weakly excites
- L16:1273 weakly excites
- L8:2689 weakly suppresses
- L22:19 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
- 方式进行
- entimes
- msgs
- 齿
- 网络传播
- .RegisterType
- 培养学生
- 昨晚
- 本科生
- issy
Suppresses
- ếc
- /site
- Eğitim
- abilité
- wann
- blanco
- الفريق
- Marker
- ever
- okia