unclear: metadata labels and closing quotes
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. 13 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 is hard to pin down. It fires most on the word "By" in a catalogue entry for who deposited a document, and on some document breaks, but these cases look unrelated. Turning it up or down did not produce 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.
-
that will give a useful input to those interested in Artificial Intelligence in education.⏎|Subjects:||L Education > L Education (General)|⏎|Deposited By:||Zaidatun Tasir|⏎|Deposited On:||
-
:⏎A = epsilon.c.l⏎Where A is the absorbance measurement at a given wavelength⏎epsilon is the extinction coefficient in M^-1.cm^-1⏎c is the concentration of the material in mol.dm^-3 (molar)⏎l
-
to large step sizes.⏎- Scott E. Fahlman: An Empirical Study of Learning Speed in Back-Propagation Networks, September 1988<|endoftext|>Cardio and weight
-
a base for cultivating other.⏎The spawn is used for transferring mycelium onto any material from which mushrooms will grow, called a substrate. There are plenty of substrates, straw, cardboard, logs, and wood chips; these are some best
-
+ΦV) and current I(t)=Imax×sin(ωt+ΦI) where Vmax, Imax are the peak values, ω is the common angular pulsation and ΦV, ΦI are the
-
⏎- Warn others when passing⏎- Keep dogs on a leash at all times for their protection, and for yours⏎- Respect all wildlife encountered on the trail<|endoftext|>In 1847, when Mormon pioneers arrived in the Salt Lake Valley
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: the current period, and the current period of the current period. In the current period. The current period refers to the current period. The current period refers to the current period. The current period refers to the current period. The current period
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Richard Feynman, and one of his most memorable quotes explains that “Science is the belief in the ignorance of the experts.”⏎Indeed, the “experts” of science are so often wrong that it makes their arrogance laughable, and today
-
study published in the journal Aggressive Behavior supports the idea that perfectionism may develop in a social group and suggests that indirect aggression triggers it. The "aggressor" talks behind a person's back<|endoftext|>You are using an outdated and unsupported browser.⏎
-
It is as if Mill began with the harm principle in mind, then concocted a description of liberty to justify it. The “devil,” in this case, lies not in the details but in the harm principle:⏎That principle is, that
-
he?⏎He delays. We know he can. No doubt. We pray he will, realizing he may not.⏎Struggling with delayed answers? Maybe those delayed answers are to
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
- L8:901 suppresses
- L9:2212 suppresses
- L11:324 excites
- L7:1893 weakly suppresses
- L12:2455 weakly suppresses
- L11:2750 weakly excites
- L11:2187 weakly excites
- L12:2672 weakly excites
- L7:1870 weakly excites
- L11:143 weakly suppresses
- L8:440 weakly excites
- L12:221 weakly excites
- L12:285 weakly excites
- L11:2085 weakly excites
- L6:1071 weakly suppresses
- L7:1267 weakly suppresses
- L7:1265 weakly suppresses
- L8:1129 weakly excites
- L2:291 weakly excites
- L11:526 weakly excites
- L3:158 weakly suppresses
- L11:2087 weakly suppresses
- L0:868 weakly excites
- L6:2985 weakly suppresses
Feeds later layers
- L15:2362 suppresses
- L14:1843 suppresses
- L14:1038 suppresses
- L14:1660 suppresses
- L16:2833 suppresses
- L15:2319 suppresses
- L20:2755 suppresses
- L15:617 excites
- L14:2766 suppresses
- L18:992 suppresses
- L14:353 suppresses
- L20:470 suppresses
- L19:1638 suppresses
- L14:1989 suppresses
- L16:1554 suppresses
- L21:765 suppresses
- L15:2787 suppresses
- L14:1558 suppresses
- L14:1839 suppresses
- L14:2786 suppresses
- L14:2653 excites
- L14:995 suppresses
- L14:696 suppresses
- L14:2947 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
- ותר
- рес
- ongan
- ButtonModule
- urses
- しかない
- ivre
- 都不敢
- evin
- vẻ
Suppresses
- 指的是
- 是指
- Ston
- "?
- Immigration
- referred
- referring
- refers
- '?
- diam