'in' in 'bear in mind' 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. 21 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
Fires on the word 'in' in set phrases like 'bear in mind', 'keep in mind' and 'with this in mind', where 'mind' is about to follow. When it was turned up, the model kept producing 'in mind' over and over.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
them fascinating. After you have mastered some fundamental activities, you’ll be able to try new ones and incorporate a brand new twist for the previous kinds. And bear in mind, it’s in no way too early to start creating new memories. >>>
-
The⏎Halacha section will primarily be a review of the lessons from the previous⏎cycles while the Aggadah section will be new material. Please bear in mind⏎that the purpose of learning the laws of Shmiras haloshon
-
forest for one use if it's going to compromise other uses? Our forests are incredibly important for harvesting but also for tourism [and] we want to always bear in mind the possible tradeoffs.⏎Tourism is a huge asset to Maine
-
? Please include any references to your response that is indicated in the Torah.⏎Answer: Before I answer your question, as a word of background, you need to keep in mind that Judaism, ever since the destruction of the second Temple in 7
-
for the business⏎- Clear goals for the management of livestock⏎The management process should involve:⏎- Reviewing these strategies regularly⏎– Making decision with these goals in mind⏎– Monitoring the health of your land, your business and the people
-
of the text, as well as the reason why the text will be of interest to the reader.⏎- Many people involved with writing hate to sketch. With this in mind, one simple piece of advice can be given: tackle the text as soon
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 am now, on the current, to be a new, which is to say, to be a man on the world's, to be a man, to be a man, to be a man, to be a man, to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
materials is creating big data.⏎Take, for example, the emergence of Moocs – massive open online courses. While some universities
-
: maintain a tree board or department, having a community tree ordinance, spending at least $2 per capita on urban forestry and celebrating Arbor Day.⏎Alison Oelschl
-
eat is the main ingredient in your diet.⏎Food is food insofar as our tastes are concerned. But, when it comes to our stomachs, and, by extension, our whole body is involved, food is more than just food. It
-
lenses and how to use them. It’s a very useful book and provides a constructive way of viewing problems.⏎A third<|endoftext|>Mecklenburg-Western Pomerania, also known as Mecklenburg-Vorpommern, is a
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
- L19:2394 excites
- L17:1368 excites
- L7:1494 suppresses
- L18:2180 suppresses
- L1:1244 weakly suppresses "of" after "board" or "off"
- L12:175 weakly suppresses
- L18:2372 weakly excites
- L12:2949 weakly excites
- L16:1297 weakly suppresses unclear: document ends and answer colons
- L9:1483 weakly excites
- L20:2186 weakly suppresses
- L20:30 weakly excites
- L5:1029 weakly suppresses
- L18:1257 weakly suppresses
- L10:1611 weakly excites
- L14:152 weakly suppresses
- L17:1691 weakly suppresses
- L2:2649 weakly excites
- L16:2418 weakly excites
- L12:2422 weakly excites
- L20:1539 weakly excites
- L9:2590 weakly suppresses
- L16:1059 weakly excites
- L10:2439 weakly excites
Feeds later layers
- L23:1788 strongly excites
- L27:1619 excites
- L23:2944 excites
- L22:2230 excites
- L26:2125 excites
- L27:1297 weakly excites
- L22:2468 weakly excites
- L27:1383 weakly excites
- L24:2218 weakly excites
- L25:23 weakly suppresses
- L24:2573 weakly excites
- L22:2909 weakly excites
- L22:2562 weakly excites 'to' after 'things' or 'factors'
- L23:2056 weakly excites
- L25:1776 weakly excites
- L23:1979 weakly excites blank before a number word
- L23:2597 weakly excites
- L23:2659 weakly excites
- L25:1008 weakly suppresses
- L26:2993 weakly excites
- L24:441 weakly suppresses
- L22:687 weakly excites
- L23:510 weakly suppresses
- L26:2348 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
- remember
- mind
- recall
- Remember
- memory
- recalling
- remembering
- Recall
- 记忆
- memory
Suppresses
- JW
- jer
- HERE
- jet
- APPER
- NewLabel
- strar
- Dion
- encer
- Crow