'guess' and 'obvious' after 'as you can see'
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. 6 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 words like 'guess', 'guessed' and 'obvious' that follow phrases such as 'As you may have guessed'. It is quiet, or pushes the other way, on the comma that closes a sentence starting with 'As the study has shown'. Turning it up did not give a clear theme.
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.
-
trying to open up the San Gorgonio wilderness so that tows and lifts for downhill skiing might be built on its hoary upper slopes. As anyone might guess, all manner of conservationists and outdoorsmen have rallied to defend the wilderness against
-
rating of 4, and Malawai has a rating of 1.3.⏎Working individually, take five minutes to briefly write down what you can surmise from the map, addressing these questions:⏎- What patterns do you see?⏎- What
-
peers were sexually active. However, they were not similarly influenced by visual depictions of sex.⏎Since two-thirds of the study's participants were women, conclusions based on gender differences could not be made. Taylor noted that other research has indicated women and
-
that is split over several lines in HTML. </BODY> </HTML>When this is displayed by Mosaic, it will look like:⏎As you can probably guess, commands are enclosed in angle-brackets <>, so that the HTML command <
-
718<|endoftext|>Women who have kids later on in life may live longer, according to the findings of a recent study.⏎Following the birth of a woman’s last child, certain measurements
-
the types of broken bones, including:⏎TransverseA bone break is considered transverse when the fracture extends across the bone.⏎SpiralAs you may have guessed, a spiral fracture spirals around the bone.⏎ObliqueAn oblique fracture
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 number of people who have been in the world of a million people. So, if there are 100 people who are in the world of a million people, each of the 100 people who have been in the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the intersectional nature of the violence faced by Black women and trans people through efforts such as #SayHerName. As the historian Keisha Blain has demonstrated, these women build on a long legacy of Black women’s activism.⏎As historians of
-
USDA data puts the variety diversity at 2,450 types.⏎The Way We Eat Apples Wastes Almost A Third⏎As the video above demonstrates, eating an apple in a circular pattern around the core
-
has been marked by pain and tragedy, and it has evidently imprinted many traumatic memories upon Indigenous communities. As Utrikesperspektiv.se has previously reported, the Canadian state is guilty of forceful assimilation practices which have had various negative
-
addressed in this book. As discussed in chapter 8, graphs at their simplest are just nodes and edges, with nodes representing things and the
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
- L3:246 excites
- L3:458 weakly suppresses 'that', 'about' or 'of' after a verb or noun
- L2:616 weakly excites
- L1:1990 weakly suppresses
- L5:1495 weakly excites
- L3:1652 weakly excites
- L2:492 weakly suppresses
- L5:967 weakly excites
- L5:316 weakly excites
- L5:1280 weakly excites
- L5:436 weakly suppresses
- L0:2044 weakly suppresses
- L4:1960 weakly suppresses
- L4:682 weakly suppresses
- L1:2577 weakly excites
- L1:1574 weakly excites
- L3:3038 weakly excites
- L4:1814 weakly excites
- L0:889 weakly excites "how do you" question phrase
- L5:2306 weakly suppresses
- L2:520 weakly excites
- L1:1580 weakly suppresses
- L4:2644 weakly excites
- L4:2414 weakly excites
Feeds later layers
- L11:1774 suppresses
- L9:841 suppresses
- L10:1159 suppresses
- L12:1632 suppresses
- L18:3010 suppresses 'to' before 'be' after considered
- L8:615 suppresses
- L8:200 suppresses
- L10:571 suppresses
- L10:609 suppresses
- L15:27 weakly suppresses
- L8:997 weakly suppresses
- L18:1519 weakly suppresses
- L8:2194 weakly suppresses
- L18:2414 weakly suppresses
- L16:1006 weakly suppresses
- L8:132 weakly suppresses
- L16:1482 weakly suppresses
- L9:272 weakly suppresses
- L15:2053 weakly suppresses
- L14:2139 weakly suppresses
- L16:2097 weakly suppresses
- L18:1690 weakly excites
- L17:2986 weakly suppresses
- L9:2247 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
- 孚
- Impact
- elp
- OrFail
- itia
- ypi
- Props
- Leipzig
- Endpoints
- Them
Suppresses
- conde
- dimension
- /posts
- 反正
- maximum
- 最大的
- факт
- =
- 最小
- /not