pronoun 'it' pointing back to a thing
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. 14 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 pronouns like 'it', 'they' or 'their' when they refer back to a thing or group mentioned earlier in the sentence. Turning it up made the text drift into unrelated lines, so the output tells us little about this 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.
-
antzen seems to have been a deacon. The members of this Lithuanian congregation were mostly of Dutch descent; they had also won some Lutherans for their church, which was taken very ill by the Prussian King. The congregation belonged
-
It states:⏎All persons born or naturalized in the United States, and subject to the jurisdiction thereof, are citizens of the United States and of the State wherein they reside.Once again the words of the Constitution identify two ways and only two ways that
-
Instead of measuring impact by land area like other studies do, theirs uses a new framework to measure the environmental impact of each farming type against the quantity of food that it produces. This revealed that by yield, high-intensity farming was less impactful than
-
have come to believe that the whole world is an enigma, a harmless enigma⏎that is made terrible by our own mad attempt to interpret it as though it had⏎an underlying truth."⏎-- Umberto Eco⏎More information about the Python
-
see: basically nothing. The town—and “town” is even sort of grand for what it was—has about 8,000 people in it. But here
-
it, technology, in the form of an automated house, is portrayed as mindless and soulless, continuing to perform its many functions for the humans who created it but are no longer there.⏎Eventually, the house (technology) is destroyed by
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 think that, maybe it's not the best way to just say that. Maybe you could say, maybe you could say that, maybe that. Maybe, maybe, maybe maybe, maybe, maybe, maybe, maybe, maybe, maybe
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
flora by means of an appropriate diet, accompanied by a probiotic supplement, offers positive effects in reducing anxiety.⏎Probiotics are microorganisms which are present in their native state in all of us from birth: they’re in the gut, the
-
so on the German side, the Enigma cipher system, the Walter propulsion system and of course the German jets, rockets and missiles are just few of the items they came up with. In these pages we'll attempt to explain most technologies relevant to
-
design and delivery of policies, programs and services that affect them.⏎How do families living remotely connect?⏎Mornington Island residents have poor quality mobile
-
that to U.S. health officials.⏎According to the Centers for Disease Control (CDC), the ALA's N-O-T program is the only smoking cessation program it rates as a "model program." In fact, the CDC does not list any
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
- L9:175 suppresses
- L8:40 suppresses
- L10:1389 suppresses
- L4:499 suppresses
- L12:2897 excites
- L13:1262 suppresses
- L12:521 excites
- L5:583 excites
- L7:2482 suppresses
- L4:1306 excites
- L8:2143 suppresses
- L13:2601 excites
- L5:2733 suppresses
- L7:2537 suppresses
- L13:1997 excites
- L7:1905 excites
- L10:2920 excites
- L6:1922 excites
- L11:2714 weakly excites
- L1:1871 weakly suppresses
- L13:2186 weakly excites
- L10:2853 weakly excites
- L13:1538 weakly excites
- L11:1245 weakly suppresses
Feeds later layers
- L16:2272 excites
- L15:396 excites
- L17:578 excites
- L17:1167 suppresses
- L17:638 excites
- L17:321 excites
- L15:714 excites
- L19:582 suppresses
- L17:314 excites
- L17:1070 excites
- L15:1226 suppresses
- L15:1131 excites
- L18:358 weakly excites
- L15:1047 weakly suppresses
- L15:26 weakly excites
- L18:1397 weakly suppresses
- L16:1702 weakly excites
- L16:577 weakly excites
- L15:1127 weakly excites
- L17:45 weakly excites
- L19:840 weakly excites
- L15:2880 weakly suppresses
- L19:2261 weakly suppresses
- L16:499 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
- ($)
- overrun
- KP
- overturn
- zf
- �
- __;⏎
- ISCO
- ($_
- Poz
Suppresses
- им
- him
- eren
- ODO
- iness
- ytt
- uyện
- igkeit
- 情况进行
- ías