mid-word fragments of "cognitive" and similar words
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. 3 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 middle piece of a word that the model has split into parts, most strongly in "cognitivists" and "cognitively". Other examples are fragments in "scholarworks" and "snarling". The pattern is not very consistent beyond that, and turning it up did not give a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
for the greatest number (Utilitarianism), it can always be questioned.⏎People can still have differences of opinion about things.⏎“Ethical naturalists are cognitivists, who believe that moral statements are propositional, and so can be
-
als hunted birds for their fashionable feathers for thousands of years and may have exploited certain plants for their medicinal properties–compelling evidence that our hominin cousins were cognitively sophisticated.⏎-<|endoftext|>During the
-
a test, so fearful of getting the wrong answers that they do not complete as many items as their peers. But test anxiety is actually quite different from a cognitively-based slow processing speed.⏎However, many observations of anxious kids suggest that the
-
ched body, weight shifted away from the fear-eliciting stimulus) accompanied by aggressive signs such as piloerection, barking, growling, snarling and biting. Dilated pupils, increased respiratory rate and a rapid heart rate
-
⏎To perform the test, the patient is instructed to stand on one leg without support of the upper extremities or bracing of the unweighted leg against the stance leg. The patient begins the test with the eyes open, practicing once or twice
-
or emotive, or do they correspond to an objective, universal goodness which we attempt to access through moral debates? What does ‘good’ mean?⏎Cognitivists – moral statements are propositional and therefore deal with verifiable facts⏎
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 have been given to that. I have been given to that. I have been given to that. I have been given to that. I have been given to that. I have been given to that. I have been given to that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, honey and bees, fruits, and the like; and uses its fore paws with a deftness almost equal to that of a monkey. It is found from Central Mexico to
-
White Fragility: Why It’s So Hard for White People to Talk About Racism by Robin Diangelo⏎“Antiracist educator Robin DiAngelo deftly illuminates the phenomenon of white fragility and “allows us to understand racism
-
meetings. They were prepared to award her the prize but Houdini protested that he needed to see for himself. At the seance, Houdini saw through the deception and called “Margery” on it, much to the annoyance of
-
attached to the museum's awkward juxtaposition with the state Division of Taxation building.⏎Beginning in 1973, the N.J. Supreme Court saw fit to interject itself into education policy and
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
- L2:1321 weakly suppresses
- L1:1476 weakly suppresses
- L2:1888 weakly excites
- L2:2292 weakly excites
- L1:2230 weakly suppresses
- L1:2186 weakly excites
- L0:2071 weakly excites
- L0:1082 weakly suppresses
- L0:1297 weakly suppresses
- L2:463 weakly excites
- L0:1941 weakly suppresses
- L1:208 weakly suppresses
- L2:715 weakly suppresses the word 'following' after 'the'
- L0:777 weakly excites
- L0:3020 weakly excites
- L2:2810 weakly suppresses
- L1:702 weakly excites
- L2:1178 weakly excites
- L1:2543 weakly suppresses
- L2:2119 weakly suppresses
- L2:1356 weakly suppresses
- L2:2367 weakly suppresses
- L0:1413 weakly excites
- L1:2915 weakly suppresses
Feeds later layers
- L4:285 suppresses
- L16:836 weakly excites
- L10:1431 weakly excites
- L4:3046 weakly excites
- L4:3071 weakly excites
- L4:2893 weakly excites the word 'means' after a subject
- L4:2976 weakly excites
- L20:1856 weakly suppresses
- L18:2934 weakly suppresses
- L19:531 weakly excites
- L6:616 weakly excites
- L6:1906 weakly suppresses
- L19:1967 weakly excites
- L6:165 weakly suppresses
- L22:2504 weakly suppresses
- L19:513 weakly suppresses
- L15:2944 weakly suppresses
- L20:1907 weakly suppresses
- L4:1444 weakly suppresses
- L4:2715 weakly excites
- L10:2075 weakly excites
- L17:1619 weakly excites
- L17:1987 weakly excites
- L8:2421 weakly 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
- olia
- XD
- unch
- 吉利
- beiter
- вел
- LowerCase
- ghan
- xed
- REPL
Suppresses
- handleError
- liği
- Matter
- .handler
- 一对
- handler
- handled
- eries
- cbc
- iner