help-seeking 'I don't know' phrases
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. 8 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 fires on phrases where someone admits they are stuck or do not know something, such as 'I'm lost', 'I don't know', or 'where to start'. Turning it up pushed the model toward writing about problems and solving them.
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.
-
is⏎I got the f(x) = x-1 part, and I got how to calculate the variance after you have the expected values, but I'm lost on other questions.⏎My questions are:⏎Where do we get if x=1
-
, Past, Future Indefinite, Present, Past, Future Continuous⏎помогите пожаааалуйста, сам не справляюсь(⏎Y. Раскройте скобки, употребляя г
-
3 (molar)⏎l is the path length in centimetres⏎i want to calculate the concentraton only after determining the extinction coeff. then hw do i determine the extinction coeff of supercoiled DNA, for example??⏎if you
-
is my weakness programming i feel an igonorant when it comes to programming my filed requires from me to take this course and i dont know anything,. sombody can help me with following program please!!:⏎Write an assembly
-
their asceticism, especially by their attachment to dhikr, the practice of remembrance of God, often performed after prayers<|endoftext|>Don’t know where to start with growing from seed? Are you worried that it will be difficult & require special
-
given to you?⏎October 3rd, 2010, 12:50 PM⏎i started the homework⏎i dont know how to finish it here what i
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 will be the one who follow the same pattern as the first problem. The problem is about the 2020 deadline 10 (10) (10) (10) (1) (11)
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
want to calculate the concentraton only after determining the extinction coeff. then hw do i determine the extinction coeff of supercoiled DNA, for example??⏎if you've your three forms in the sample i
-
are the part of same Microsoft exchange server.⏎How do I know if a recalled communication was successful in outlook?⏎Make sure you have to tick the following option when recalling the email. Outlook will send you a
-
top percentage of the class, or when a student drops to the lowest percentages of the class. Select the stream alert to view the data.⏎In the list where your name appears, select Grades. On the Grades page, select the cell that displays
-
policy, in a simulated domain.⏎the application examples in programming are very good and i loved how rl gets closer and closer to how a living being thinks. if you take a course in audit mode, you will be able to see most course materials
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
- L3:255 excites
- L3:148 suppresses
- L7:676 weakly suppresses
- L7:1076 weakly suppresses
- L7:61 weakly excites
- L7:30 weakly suppresses
- L6:1167 weakly suppresses
- L7:563 weakly excites
- L7:1981 weakly excites
- L2:1350 weakly suppresses
- L6:2189 weakly excites
- L1:599 weakly suppresses
- L1:2128 weakly suppresses
- L5:1466 weakly suppresses
- L4:413 weakly excites
- L6:17 weakly suppresses
- L6:16 weakly excites
- L3:166 weakly suppresses
- L6:637 weakly excites
- L6:2437 weakly suppresses
- L3:213 weakly suppresses
- L1:887 weakly excites 'be' after 'might', 'also' or 'only'
- L7:1698 weakly suppresses
- L1:566 weakly excites "addition" in "In addition"
Feeds later layers
- L9:373 excites
- L13:1309 suppresses
- L9:276 suppresses
- L9:2680 excites
- L9:655 weakly suppresses
- L11:2376 weakly excites
- L10:1112 weakly excites
- L10:2808 weakly excites
- L9:585 weakly excites
- L11:561 weakly suppresses
- L10:1342 weakly excites
- L14:28 weakly excites
- L12:863 weakly excites
- L11:2350 weakly excites
- L11:2791 weakly excites
- L11:2414 weakly suppresses
- L13:888 weakly excites
- L10:3048 weakly suppresses
- L9:184 weakly suppresses
- L9:757 weakly suppresses
- L19:1547 weakly suppresses
- L9:1207 weakly excites unclear: mid-sentence tokens in web text
- L14:1218 weakly suppresses
- L20:2629 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
- يد
- roid
- factors
- typeId
- オス
- ueil
- leshoot
- /sources
- 装修
- isors
Suppresses
- .onActivityResult
- 首发
- upe
- thờ
- setTitle
- ainen
- atasets
- ginger
- -relative
- .setPrototypeOf