'long' in 'in the long run'
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. 22 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 the word "long" in the set phrase "in the long run" or "in the long term". When turned up, the model kept repeating "long term term", so it clearly pushes the model toward that phrase.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
habitat may become unsuitable in just 50 years. Populations at the south-western edge of the species’s distribution may be the most viable in the long term.⏎Worthen’s Sparrow Spizella wortheni is currently listed
-
your diet. Although processed food has an undeniable convenience factor and longer shelf life, they typically contain unnecessary additives and chemicals that can be detrimental to your health in the long run.⏎While not always the budget-friendly option, there are many benefits to eating
-
days by a fraction, hardly had an impact on the planet in the long run. In fact, scientists have a hard time spotting the effect of even bigger
-
HG symptoms early in pregnancy can make you less sick in the long run and can decrease recovery time. Because HG is caused by many factors that vary
-
which is more resilient and cost-effective in the long term. Local governments need systems that enable them to incorporate the health of natural resources
-
have a much bigger impact on the budget deficit in the long-run than a cut to basic science research or education can. By complementing the
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: A state's feds are just a part of the state's rule, so they can't be a country. Therefore, his fads are just a part of the state's fads, which are just a part of the state's
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ons, cartilage, muscle, nerve and fat.|⏎|Myeloma||A type of blood cancer that hits the plasma cells of bone marrow. Myeloma cancer cells make abnormal antibodies.|⏎|Leukemia||A type of
-
times as likely to have multiple myeloma (a malignant bone tumor).(3) Below is a tabulated version of
-
ness and irritation, and subsequent spread along the natal cleft is common.⏎Candidosis, Genital/Perigenital, Kolpitis Candidomycetica, Vulvitis Candidom
-
has spread to the lining surfaces of the peritoneal (abdominal) cavity from ovarian cancer, primary colorectal cancer, appendiceal cancer, or mesothelioma and pseudomyxoma per
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
- L18:1641 excites the word 'long' in phrases
- L16:108 weakly excites
- L1:2130 weakly suppresses
- L11:2456 weakly suppresses
- L10:378 weakly suppresses
- L19:2850 weakly suppresses
- L8:2620 weakly excites
- L16:2622 weakly suppresses
- L10:22 weakly suppresses
- L3:37 weakly suppresses
- L18:38 weakly excites
- L1:1523 weakly suppresses
- L18:133 weakly excites
- L13:2805 weakly excites
- L20:150 weakly excites
- L0:2912 weakly excites
- L10:323 weakly suppresses
- L12:402 weakly suppresses
- L20:183 weakly excites
- L0:1593 weakly suppresses
- L11:1688 weakly excites
- L16:1085 weakly excites
- L21:280 weakly suppresses
- L1:1654 weakly excites
Feeds later layers
- L25:2278 excites
- L23:2238 weakly excites
- L24:712 weakly excites 'long' in 'long-term' and 'long-lasting'
- L27:1426 weakly excites
- L23:2927 weakly suppresses
- L25:2934 weakly suppresses
- L25:1886 weakly excites
- L23:2280 weakly suppresses
- L27:1178 weakly excites
- L24:2966 weakly excites
- L23:2629 weakly suppresses
- L26:265 weakly excites
- L24:1839 weakly excites
- L26:2777 weakly excites
- L23:1270 weakly suppresses
- L27:2540 weakly suppresses
- L25:488 weakly suppresses
- L23:2088 weakly suppresses
- L24:2692 weakly excites
- L23:2331 weakly suppresses
- L23:2613 weakly suppresses
- L26:2046 weakly suppresses
- L23:459 weakly suppresses
- L24:2342 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
- -term
- term
- term
- _term
- TERM
- Term
- _TERM
- Term
- terme
- .term
Suppresses
- 菲尔
- stras
- spa
- ngân
- bane
- irates
- ever
- gba
- 妻
- SEL