unclear: 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
It fires on small connecting words in the middle of a sentence, like 'a', 'of', 'in' or 'up', where the phrase is clearly about to continue. It is hard to name one clear topic. When it was turned up, the model's writing became repetitive and wandered off, with no clear theme.
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.
-
(IORefs/STRefs for example).⏎> One way to avoid that is for the leaves to refer to their objects⏎> through a Map. Then changing a leaf object requires only that the value⏎> associated with key represented by the leaf
-
en.]⏎Orthography: The Cleasby & Vigfusson book used letter ö to represent the original Old Norse vowel ǫ. Therefore, bröstuliga may be more accurately written as brǫstuliga.⏎Possible run
-
life on Mars, and before tech billionaires started threatening to send people to space.⏎So if NASA’s Planetary Protection Officer—there is one—is going to live up to the coolest job title in the solar system, the space agency is going to
-
, from generation to generation. - (C.7323)<|endoftext|>EffectivenessA blood pressure reading consists of two numbers -- for example: 120/80. The top number is known as the systolic blood pressure
-
think about the matter of teen sex. What, she asked, are the real consequences of sexual precocity?⏎To fully understand Meier’s motivation, we must rewind to 1996, when
-
around which the entire northern sky appears to turn. That is because Polaris is located more or less above the northern axis of the Earth, and the wheeling of the stars across the dome of night is really due to Earth’s turning, after
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: and I want to share my thoughts in the case of the most powerful effort of my greatest efforts and the most powerful effort of my most powerful effort of my and of the most powerful effort of my most powerful effort of my and of of my
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
outcomes between regular and advanced economics classes. Finally, the authors find that students in advanced economics classes significantly outperform students in regular classes.⏎If you experience problems downloading a file, check if you have the proper application to view it first. In case
-
|⏎|Estimated Time||50 minutes (or a series of minilessons)|⏎When teaching a lesson on descriptive writing, students may write descriptively in the moment,
-
a historic event of such significance. Here, you will find useful tips, guidelines and possible topics that might be interesting to write in your paper on history.⏎Writing a paper on history requires you providing lots of information and historical details within the given amount
-
Specific, Measurable, Attainable, Relevant, and Time Bound. Setting a goal and seeing it through to completion requires careful planning and consideration.⏎When setting a S.M.A.R.T goal, it is important to be specific about what you
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
- L14:1288 strongly suppresses
- L14:2194 suppresses
- L13:943 excites
- L14:1608 suppresses
- L14:1463 suppresses
- L13:1071 excites
- L13:958 excites
- L13:1711 excites
- L13:1434 suppresses
- L13:359 suppresses
- L12:1660 suppresses
- L13:401 excites
- L13:487 excites
- L13:1343 suppresses
- L14:2325 suppresses
- L14:2999 suppresses
- L13:891 suppresses
- L13:516 suppresses
- L14:1526 excites
- L14:1193 suppresses
- L12:210 suppresses
- L14:2118 suppresses
- L14:924 suppresses
- L14:186 suppresses
Feeds later layers
- L16:1265 suppresses
- L17:1502 excites
- L18:1187 excites
- L17:1640 excites
- L16:1342 excites
- L17:2496 suppresses
- L16:2869 excites
- L16:1802 excites
- L18:658 excites
- L18:1414 excites
- L16:411 excites
- L20:210 excites
- L16:1587 suppresses
- L16:397 excites
- L18:1003 excites
- L16:320 excites
- L16:1719 suppresses
- L16:1646 excites
- L17:1746 excites
- L21:881 excites comma after 'if' clause in prose, braces in code
- L19:1475 weakly suppresses
- L16:669 weakly suppresses
- L16:931 weakly suppresses
- L17:200 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
- uling
- dạ
- olars
- usher
- athing
- '<%=
- 系统的
- ◢
- 天涯
- putc
Suppresses
- omon
- sav
- sez
- 麦克
- subclass
- regor
- pecially
- �
- logue
- 小康