"but" in "not only X, but" 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. 17 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 lights up on the word "but" when a sentence has the pattern "not only this, but also that". It is a very consistent trigger, but turning it up or down only produced loose or broken text, so the steering does not show a clear effect.
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.
-
. Active student participation is not only encouraged, but required.⏎At some institutions, Logic & Proofs has been offered as a fully
-
you providing lots of information and historical details within the given amount of words and pages. It is essential to include not only related to a certain historical event information, but also to show a coherent, profound manner of analyzing. You are expected to demonstrate
-
women in newspaper work not only to write the news, but to make it.”⏎Marjory Lang, Women Who Made the News (2
-
you must understand and apply some crucial National Electrical Code rules and regulation. Following these rules to the letter will not only assure you that your work will pass inspection, but that it will be a safe installation as well. Improperly installed recessed
-
your phone and go back to your routine. But this might change soon. Now, your shoes can not only count your steps but also study them.⏎“This means that now you can understand how your feet are being
-
also being seen adopting Pilates exercises. The coaches and athletes are not only using pilates to benefit from improving core strength and flexibility, but also to prevent injuries. Pilates aids the athletes capacity in their field. If
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: this this for 8 years 36. 3 6 3 .... 5 ... ... 5 6
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. In so doing, you might be aiding your youngster to be a far better thinker and<|endoftext|>Take any point P inside an equilateral triangle. Draw PA, PB and PC⏎from P perpendicular to the sides of the triangle where A, B
-
letter words⏎6 letter words⏎5 letter words⏎4 letter words⏎3 letter words⏎2 letter words⏎Images for ZEATINSLoading...⏎SCRABBLE is the registered trademark of Hasbro and J.W. Spear & Sons Limited
-
neck etc. of daevic beings (see DĒW), and mentioned manaoθrī- as the corresponding ahuric word (AirWb., cols. 530, 1126; cf.
-
activation of the kinases leads to the initiation of downstream restoration pathways. Likewise, in response to Ataluren pontent inhibitor solitary strand breaks in DNA (SSBs), PARP1 as well as the MRN complicated initiate repair from
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
- L1:4 strongly excites
- L0:825 strongly suppresses
- L2:1855 strongly excites
- L4:1319 strongly excites the 'B' in vitamin B
- L16:260 excites
- L16:374 excites
- L10:201 suppresses the letter B in vitamin names
- L7:520 suppresses
- L6:1357 excites
- L16:75 suppresses
- L7:2095 excites
- L10:887 suppresses the B in vitamin B names
- L16:134 excites
- L6:328 suppresses
- L16:160 excites
- L16:827 suppresses
- L0:1638 suppresses
- L16:215 suppresses
- L16:671 excites
- L5:138 excites
- L16:249 excites
- L15:129 suppresses
- L16:1079 suppresses "by" after a past participle
- L16:1358 excites
Feeds later layers
- L20:256 strongly excites
- L20:174 suppresses
- L19:370 suppresses
- L18:328 suppresses
- L18:242 excites
- L18:198 suppresses
- L18:474 suppresses
- L18:404 suppresses 'get', 'go' and 'got' after a subject
- L18:344 suppresses
- L19:583 excites
- L18:1493 suppresses
- L18:205 suppresses
- L18:102 suppresses
- L18:385 excites
- L18:936 suppresses
- L18:365 suppresses
- L20:211 suppresses
- L19:197 suppresses
- L22:2245 excites
- L20:183 suppresses
- L18:103 suppresses
- L19:569 suppresses
- L21:706 excites
- L19:754 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
- heading
- gest
- venue
- trand
- urst
- hev
- кров
- 卖给
- tester
- 可惜
Suppresses
- yum
- igue
- wort
- inic
- irty
- eus
- uits
- WG
- еть
- minut