comma after a transition word
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. 12 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 on the comma that follows a linking word at the start of a sentence, such as "In turn,", "Instead,", "Furthermore," or "Similarly,". Turning it up made the model write repetitive, unrelated text, so the steering does not confirm the pattern.
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.
-
une Function, 2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20potent%
-
age of first-time mothers since the age of a woman’s first birth has a bearing on the total number of births she may have within her lifetime. In turn, this has an impact on family size and future increase in the overall population. In
-
diagnosed or undiagnosed depression often use drugs or alcohol as a form of self-medication. This then changes the brain structure and makes depression worse. Or, someone without depression starts using, and the changes in their brain cause depression.⏎In
-
capacity for this quality. Other experts disagree and argue that compassion isn’t the root cause of fatigue or stress. Instead, they say a general lack of tolerance<|endoftext|>Pack of 8 hanging succulents
-
be breakable even with today's computers, in which case the elliptic-curve algorithm will still provide the best security that today’s technology can offer. Alternatively, if the post-quantum algorithm turns out to be secure then it'll protect
-
building’s demand for electricity from the grid.⏎- Passive systems: The building design includes many passive features such as maximizing natural daylight and ventilation from the outdoors and, low-<|endoftext|>Five Questions about Long-Term Care⏎1. What is long
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: and I think, "and I have, and I have, and I will, and I will, and I will, and I will, and I will, and I will, and I will, and I will, and I will
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, a renowned American poet, has captivated readers with her profound exploration of nature and the human experience. Her poems often evoke a sense of wonder and contemplation, inviting readers to reflect on their own lives. In this argumentative essay, we
-
was decision-making within organizations and is best known for the theory of “bounded rationality” and the theory of “satisficing.” He applied psychology to statistics, operations, and decision-making scenarios to help define the process that people go through when
-
3. Symbolism: [Analyze the symbolic elements in the poem and their deeper meanings]⏎These devices enhance the reader’s understanding and emotional connection to the poem, making it a powerful piece of literature.⏎Impact on Readers⏎Mary Oliver’s poems
-
grow uncontrollably, forming a tumor. Early detection is crucial for successful treatment, and dogs have shown incredible abilities in detecting breast cancer at its early stages. These
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
- L1:702 weakly excites
- L9:1952 weakly excites
- L6:1115 weakly suppresses
- L10:58 weakly excites
- L10:80 weakly excites
- L9:23 weakly suppresses
- L5:3066 weakly suppresses
- L11:223 weakly suppresses
- L9:103 weakly excites "as" or "since" giving a reason
- L4:244 weakly suppresses
- L5:2656 weakly excites
- L6:234 weakly excites
- L5:1511 weakly suppresses
- L10:39 weakly excites
- L6:2139 weakly suppresses
- L4:50 weakly excites
- L10:1523 weakly suppresses
- L8:311 weakly suppresses
- L1:2051 weakly suppresses
- L0:2593 weakly excites
- L9:618 weakly excites
- L10:1534 weakly suppresses
- L4:265 weakly excites
- L7:910 weakly suppresses
Feeds later layers
- L19:2741 excites
- L16:797 suppresses
- L17:1557 excites
- L17:2978 excites
- L13:2687 excites
- L15:1370 suppresses
- L16:2451 excites comma after an introductory generalization
- L21:639 weakly excites
- L14:1237 weakly suppresses
- L14:1158 weakly suppresses 'to' starting a purpose clause
- L18:391 weakly suppresses
- L20:660 weakly excites
- L15:3044 weakly excites
- L19:2085 weakly excites
- L21:721 weakly excites
- L19:433 weakly excites
- L21:191 weakly suppresses
- L18:956 weakly excites
- L16:528 weakly excites
- L24:1432 weakly excites
- L15:27 weakly excites
- L17:1495 weakly suppresses 'without' introducing a contrast
- L19:419 weakly excites
- L22:689 weakly excites comma closing an intro clause
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
- 可以获得
- 请联系
- operations
- -eyed
- уется
- みると
- Operational
- ationale
- обра�
- BaseEntity
Suppresses
- odor
- crollView
- 犁
- range
- ramid
- ulk
- licant
- //!
- uforia
- erty