preposition 'in' after a noun
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. 20 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 fires mostly on the word 'in' when it follows a noun or verb, as in 'involved in' or 'in our everyday lives'. Turned up, the model got stuck repeating the word 'that', which is a different thing it pushes toward.
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.
-
the comfort of your own home, at your computer. With identity theft posing a real threat, though, keeping financial data private requires that consumers be proactive in the way that they approach online security.⏎The Rising Wave of Identity Theft⏎According to the
-
many students, these choices remain a reality for survival at the Institute. MIT students too often sacrifice sleep, exercise, social activity, and other forms of personal care in order to enhance their educational experience. The consequences of this lifestyle imbalance can be far
-
000 words— past and present—from across the English-speaking world. As a historical dictionary, the OED is very different from those of current English, in which the focus is on present-day meanings. You’ll still find these in the
-
with the FDA. It must continue all the way to our dinner tables, which means that we, as consumers, need to know and practice proper food safety habits in our everyday lives. Here are
-
of animal welfare advocates.”<|endoftext|>People that live in islands still use water transport as a means of transportation. People that rely on fishing for their livelihood use water transport in fishing. A boat is widely used in sea transport. A boat is more flexible
-
“any breeder, public or private, the opportunity to send germplasm they wish to have evaluated. The information we generate is useful to the breeders in knowing the strengths and weaknesses of potential cultivars. The data is also critical in
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: to use a todi to make a todi to do a todi to do ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
treat each other.⏎Intervene Quickly & Briefly⏎Many adults tell me that when they witness cruelty in their classrooms, they freeze up and don’t know what to do or say. I tell them that the most effective way to intervene is
-
9)⏎practice, or doing. Reality is manifested in concrete situations.⏎coming to terms with things by studying words individually.⏎choosing what perspective to take in a situation; the unveiling of an essential meaning.⏎approaches in
-
verification code via phone call⏎- SMS – the user can receive a text message containing the verification code or password reset link⏎When it comes time to decide which method(s) to use for
-
through all of it.⏎So what’s the key? Goals have to be actionable. That is the only way to make sure they will know what steps to take, and don’t waste valuable time spinning their wheels.⏎Does your teen
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
- L18:2666 suppresses
- L17:2582 excites
- L18:2740 suppresses
- L19:1235 suppresses
- L19:2621 suppresses
- L18:1389 excites
- L18:2031 suppresses
- L18:1617 weakly excites
- L17:1282 weakly excites
- L17:2783 weakly suppresses
- L15:2692 weakly excites 'to' after a noun, before a purpose verb
- L14:2528 weakly suppresses
- L3:50 weakly excites
- L19:2737 weakly excites
- L13:164 weakly excites
- L7:1152 weakly suppresses
- L17:1368 weakly suppresses
- L19:945 weakly suppresses
- L17:2695 weakly suppresses
- L19:3040 weakly excites
- L9:1872 weakly suppresses
- L17:1884 weakly suppresses
- L18:2575 weakly excites
- L7:1024 weakly excites
Feeds later layers
- L22:1502 suppresses
- L24:959 suppresses
- L21:1550 excites
- L23:830 excites
- L27:616 excites
- L22:892 excites
- L26:586 excites
- L22:631 excites
- L22:2014 excites
- L26:408 suppresses
- L22:1186 excites
- L21:2876 weakly excites
- L23:382 weakly suppresses
- L21:2711 weakly suppresses
- L24:381 weakly suppresses
- L23:914 weakly excites
- L23:2514 weakly excites
- L21:1422 weakly excites
- L22:1616 weakly excites
- L21:639 weakly excites
- L23:298 weakly excites
- L21:1851 weakly excites
- L22:181 weakly excites
- L21:1521 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
- that
- that
- dass
- →that
- rằng
- że
- _that
- sofar
- bahwa
- (that
Suppresses
- ki
- udio
- udios
- to
- quets
- .to
- ilion
- 提出了
- icious
- riet