first person 'I' or 'my' in speech
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 fairly consistent: several of them 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 on the word 'I' or 'my' when a person is speaking or writing about themselves, especially inside quotes. It is one of the strongest neurons we looked at. Turning it up did not make the writing clearly first person, though.
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.
-
picture of.⏎"God," he said.⏎"Why Johnny," she said, "you can't do thatbecause nobody knows what God looks like."⏎"They will when I'm finished," he assured her.⏎That story presents concretely, almost blatantly
-
2019, the station will operate on renewable natural gas (RNG) produced as a byproduct of wastewater treatment.⏎“Our future is green, and I’m very proud that we’re bringing the first natural gas vehicles to the City’s
-
Department and Flint Hills Resources.⏎“The state’s response teams responded pretty quickly," said Newstead. "I don’t know exactly how much, they put thousands and thousands of feet of absorb
-
10."⏎Others welcome the report. "The estimate is transparent and conservative," says particle physicist Georg Weiglein at the University of Durham, UK. "I ...<|endoftext|>The Diary of Anne Frank Essay written by l
-
and game play.⏎The app is available to download for free.⏎or visit myspira.com to download the app<|endoftext|>Section 8: The Church⏎Week
-
the IDE as well as more adept at program creation.⏎Go no further if you're looking for the course that will guide you into the world of programming! With my course I can guarantee that the knowledge you learn will help
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: climate change and the increase of the average temperature, which is 2 degrees for the purpose of designing a strategic plan. But there is a need to ensure the quality of a strategic plan. Now, we have a dynamic, flexible, and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and create models to explore how neural circuits are organized, and how they give rise to behaviour and cognition — among the deepest mysteries in neuroscience. Ultimately, said Markram, the HBP would even help researchers to grapple with disorders such as Alzheimer
-
not land a steady job. A handful of boys from up and down Senegal’s coast. The captain, Mamour Ndiaye, who always goes to sea wearing the same tank top in the pattern of the
-
44 g for France. France has no<|endoftext|>Lynne from Tennessee writes:⏎I’m trying to learn Windows 8.1 and it is really confusing me. Can you explain what: an App is and do I really need it?
-
ulators with ideas and advice to solve formulating challenges. To submit a question, e-mail FormulateThis@allured.com.⏎Question: I’m trying to understand cosmetic emulsion inversion. What actually happens inside the emulsion at the phase borders
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
- L5:52 strongly suppresses
- L6:200 strongly excites
- L7:1034 strongly excites
- L5:771 strongly suppresses
- L7:834 excites
- L4:1855 excites
- L1:1063 excites
- L5:1028 suppresses
- L7:1412 excites
- L7:555 suppresses
- L6:214 suppresses
- L4:1268 suppresses
- L7:2429 suppresses
- L6:410 weakly suppresses
- L6:1140 weakly excites
- L4:832 weakly excites
- L4:1477 weakly excites
- L7:169 weakly suppresses
- L0:1833 weakly suppresses
- L2:853 weakly excites
- L4:499 weakly suppresses
- L7:1004 weakly excites
- L7:67 weakly excites
- L5:151 weakly excites
Feeds later layers
- L20:1058 suppresses
- L11:1002 excites
- L18:1025 suppresses
- L19:2124 excites
- L17:2086 excites
- L11:1844 excites
- L12:1662 excites
- L9:919 excites
- L22:1065 excites
- L16:2810 suppresses
- L9:652 suppresses
- L25:2082 excites first-person 'I' starting a clause
- L10:419 excites
- L13:2559 excites
- L20:1110 suppresses
- L11:1741 excites
- L25:1541 excites
- L10:2693 excites
- L10:2504 weakly excites
- L9:885 weakly suppresses research and study wording
- L11:2798 weakly suppresses
- L10:1679 weakly suppresses
- L16:3065 weakly excites
- L24:1134 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
- 妻子
- 老婆
- 我就
- ible
- ificate
- assignments
- ivant
- esposa
- 建议
- 我自己
Suppresses
- .Scan
- eds
- elkaar
- dq
- _exam
- strerror
- ແ
- matter
- 鲷
- -Fi