unclear: word fragments like 'limpet' and 'lim'
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. 2 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 is unclear. It fires on pieces of words, most often the 'limpet' sea creature and word starts like 'lim' and 'kelp', and it is weak. Turning it up or down did not produce an obvious change.
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.
-
better and more robust vehicles—such as bicycles, boats, and race cars—in the not-so-distant future.⏎"It's about translating design principles found in limpets to form structures that are strong, yet light," says study leader Asa Barber
-
contain—which makes them weaker than smaller, less flawed, structures. (Also see "Watery Gecko Grip Could Lead to Stickier Tape.")⏎Not so for limpets.⏎"Limpets are clever because they use mineral fibers below a particular size,
-
instance, racing bikes and race cars use light, strong composite materials to cope with stress, and a limpet-tooth-inspired design could help engineers build a stronger chassis that doesn't sacrifice speed
-
avitreal injection of pharmacologic agent.⏎Courtesy of Stephen J. Kim, MD.⏎Injections can be accomplished safely 3–4 mm posterior to the limbus. Commonly used methods for anesthesia administration before intravitreal injections include pledgets
-
, which cruised in the Baltic made no attack on German territory, and the immunity of such places as Kiel and Danzig was, no doubt, due to the formidable batteries which guarded the
-
flowers on hand for one vase, use two, or even three. Do not crowd flowers.⏎-<|endoftext|>Pacific ecosystem: An ochre sea star makes a kelp forest home in Monterey Bay, California.⏎Image courtesy of Kip F. Evans
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 thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
agua is the highest peak in both the Western and Southern Hemispheres.⏎McKinley, North America⏎At the Denali National Park in the US state of
-
graduating in 1952. One year later he married Louise Jones, an educator and activist in her own right. The couple has two daughters.⏎McKinney served as pastor of Olney Street Baptist Church in Providence, Rhode Island between
-
.⏎Our research is ongoing and aims to best understand these elusive creatures in order to best manage them and the essential ecological processes they provide.⏎Ms Jeannine McManus is the research and field manager. She is currently doing her PhD with the
-
J.B. Kaufman, who wrote the exhibit’s catalog, called the film “the culmination” of Disney’s previous work on short, animated films featuring Mickey Mouse, among other characters. “It established a gold standard for the craft of making
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
- L0:1301 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L0:1320 weakly suppresses
- L1:1029 weakly excites
- L0:1075 weakly excites
- L1:1464 weakly excites
- L1:189 weakly suppresses
- L0:6 weakly excites
- L0:2069 weakly excites
- L0:1 weakly suppresses
- L1:1674 weakly excites
- L0:1646 weakly excites
- L1:1959 weakly suppresses
- L1:331 weakly suppresses
- L1:194 weakly suppresses
- L1:2922 weakly excites
- L1:146 weakly excites
- L0:2002 weakly excites
- L1:1460 weakly excites
- L0:611 weakly suppresses
- L1:2111 weakly excites
- L1:535 weakly suppresses
- L0:799 weakly suppresses
Feeds later layers
- L11:1982 weakly suppresses
- L10:2704 weakly suppresses
- L5:2646 weakly excites
- L7:2342 weakly suppresses
- L6:2611 weakly suppresses
- L11:2567 weakly excites
- L19:1265 weakly suppresses
- L23:2708 weakly suppresses
- L17:2068 weakly suppresses
- L5:1880 weakly excites
- L12:212 weakly excites
- L16:2822 weakly suppresses
- L17:361 weakly suppresses
- L5:2188 weakly suppresses
- L11:1764 weakly suppresses
- L5:2521 weakly excites
- L15:197 weakly excites
- L3:2642 weakly suppresses
- L4:576 weakly suppresses
- L15:2654 weakly suppresses
- L10:2067 weakly suppresses
- L24:962 weakly suppresses
- L13:2831 weakly suppresses
- L7:2399 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
- ırl
- tures
- inis
- atoria
- ǐ
- icap
- 社会资本
- inea
- Fatal
- Neither
Suppresses
- 才知道
- 활
- silence
- annunci
- 笑了
- ám
- dess
- allowed
- 蟆
- eração