words after 'as though' or 'as if'
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. 7 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 the word right after comparison phrases like 'as though' or 'as if', and on a few related words like 'also' and 'that' in advice sentences. It is a loose pattern. Turning it down or up changed the model's topic without a clear link to these phrases.
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.
-
book:⏎The two entities that were supposed to write the regulations, the Pay Board and the Price Commission, wrangled and dithered. When it looked as though they were going to miss a crucial deadline
-
obvious effect on prospective memory, the study’s authors recommend changing the prevailing view that the allele has only subtle, often undetectable effects on cognition. The findings also supplement previous discoveries of how the allele also is linked to problems in epis
-
one per student)⏎-four large pieces of paper or poster board, each with one of the following statements written in large letters: “Strongly Agree,” “Agree Somewhat,” “Disagree Somewhat,” “Strongly Disagree”⏎Activities
-
cool at 5 degrees C or below can prevent germs from growing and multiplying.⏎2. Ensure Food is stored at the right temperature⏎It is also recommended that you do not store warm or hot food items directly into the fridge. The sudden
-
without debugging (Ctrl+F5)⏎- For any runtime error, you can debug the program as follows:⏎- Select Debug >> Start debugging (F5)⏎- Please note that during debugging, the program will show and close the console display immediately.⏎
-
left is both a __________________ and __________________.⏎Your brain also solves mysteries in language. Imagine you find the cryptic note to the right on a food container. Good thing your brain needs only the first and last letter of a word to be in the
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: "What is the meaning of the line in the middle of a circle in the context of the problem, and how does it relate to the problem of the time of the day?" I need to think about the problem of the time of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
as a castaway among the native inhabitants of pre-colonial Cuba.⏎María Estrada (the surname is given as Destrada or Estrada in some sources) was born in Seville, although her father came originally from northern Spain
-
that are called whenever⏎a child is launched or terminated. (The Win32 version of Apache uses multithreading⏎rather than a multiprocess model, but as of this writing modules are not given⏎a chance to take action when a
-
to make an image of a 1GB drive, and store it on another hard drive in the same PC.⏎Note: always be cautious when using dd, as it’s very easy to completely wipe out a drive, as we will show later
-
irth: Or Pills to Purge Melancholy’ by Thomas D’Urfey in 1719 (though ‘Wit and Mirth’ was originally published in<|endoftext|>This week's mental
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
- L6:1086 weakly excites
- L2:472 weakly suppresses
- L5:302 weakly suppresses
- L0:2814 weakly excites
- L5:1710 weakly excites
- L6:2211 weakly suppresses
- L6:2040 weakly suppresses
- L6:966 weakly excites
- L5:1349 weakly suppresses
- L1:1018 weakly suppresses
- L5:1914 weakly excites
- L6:2439 weakly excites
- L6:1883 weakly suppresses
- L4:632 weakly excites
- L2:335 weakly excites
- L4:63 weakly excites
- L1:777 weakly excites
- L5:887 weakly suppresses
- L4:1736 weakly suppresses
- L3:2046 weakly excites
- L1:1528 weakly suppresses
- L5:1451 weakly suppresses
- L2:3021 weakly suppresses
- L4:570 weakly suppresses
Feeds later layers
- L8:362 suppresses unclear: evaluative or explanatory phrase words
- L8:739 excites
- L8:1723 suppresses
- L8:480 suppresses
- L13:2748 excites
- L10:2511 suppresses
- L9:118 excites
- L8:1950 suppresses
- L12:118 suppresses
- L10:1487 excites
- L8:2329 excites
- L11:1628 weakly suppresses
- L8:1291 weakly excites
- L9:1943 weakly suppresses
- L8:597 weakly excites
- L12:376 weakly suppresses
- L8:44 weakly excites
- L8:510 weakly excites
- L10:1123 weakly excites
- L12:861 weakly suppresses
- L13:385 weakly excites
- L10:1310 weakly suppresses
- L8:50 weakly excites
- L8:91 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
- ozo
- 那你
- aser
- 蹼
- roupe
- pomocą
- Rosenstein
- ’est
- rase
- 报记者
Suppresses
- *
- *t
- "*
- че
- σ
- icum
- schedules
- *s
- overhead
- *n