'and' closing a list of items
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. 17 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 "and" when it joins two or more names or things, like "X and Y" in a list. When it was turned up, the model started stringing together "and and and" over and over.
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.
-
, of which only three have survived: the⏎Auberge de France, the Auberge d'Angleterre and the Auberge de Provence and⏎After the foundation of Valletta in 1566, the
-
2008's major hurricanes, Gustav and Ike, as well as other powerful storms, such as 2005's Katrina and Rita and 2004's Ivan.⏎API plays two primary roles for the industry
-
well levels of arsenic, cadmium, manganese, and lead and birth defects prevalence. The studies used a statewide database of private well contaminants collected by
-
North American) family.⏎Where Do Anchovies Come From?⏎Anchovies are saltwater forage fish, native to the Mediterranean and the Black Sea and thus very popular in the local cuisine.⏎Similar to herring, anchovies run
-
iting can refer to criminally duplicating currency, documents such as passports, birth certificates, drivers’ licenses or deeds, or even goods such as clothing or accessories and representing them as being brand name products. Counterfeiting of money is a federal
-
name suggests, yellow-brown in colour. In it’s early stages, it is often mistaken for sand or dirt or a stain on the pool sides or floor and unlike green algae though, mustard algae is not slimy.⏎Mustard
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 idea of a better understanding of the world, which would allow us to better understand ourselves. Then, with that understanding, we would be better able to act, with that insight, we would have a better sense of ourselves. 这
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
's global environment for promoting peace and non-violence among mankind."<|endoftext|>Processing nodes in a graph one at a time, usually in
-
and dimension to an outfit, and can be used to create a sense of contrast or harmony. When choosing textures, it is important to consider the season and occasion, as well as your personal preferences. You can also experiment with layering different textures
-
, other sources of individual variation, or environmental conditions.⏎You have requested a machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Neither BioOne nor the owners
-
of education, from kindergarten to college and beyond. Sometimes, publishing companies hire curriculum developers when they want to update textbooks. These professionals need excellent research and fact-checking skills, and they should also understand what learners need.⏎3. Product Testing (
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
- L10:74 excites
- L13:557 suppresses
- L16:376 suppresses 'or' joining alternatives
- L13:1296 suppresses
- L4:57 suppresses
- L9:92 suppresses "and" or "or" joining two items
- L13:131 excites 'and' joining parallel phrases
- L10:1643 suppresses
- L16:2432 suppresses
- L14:1116 weakly excites 'and' inside lists
- L16:1669 weakly excites
- L13:1066 weakly excites
- L13:1132 weakly excites
- L5:1577 weakly suppresses
- L11:1005 weakly suppresses
- L7:377 weakly excites
- L2:2124 weakly suppresses
- L13:1081 weakly suppresses
- L14:1746 weakly suppresses
- L4:62 weakly excites
- L16:2996 weakly suppresses
- L5:1861 weakly suppresses
- L1:1644 weakly suppresses
- L16:3017 weakly suppresses
Feeds later layers
- L24:205 strongly excites
- L26:347 excites the 're' in self-reliant and similar
- L19:1245 excites comma opening a list of parallel items
- L27:289 excites
- L18:1856 suppresses
- L25:6 suppresses
- L18:1549 suppresses
- L26:65 suppresses
- L21:739 excites 'both X and Y' pairing words
- L22:632 excites
- L18:1655 excites 'and' joining two clauses
- L23:379 excites
- L18:957 excites
- L18:870 suppresses
- L20:2386 suppresses
- L18:1402 excites
- L20:1009 excites
- L20:75 suppresses
- L18:1207 excites
- L18:486 suppresses
- L21:2699 weakly suppresses
- L24:3054 weakly excites
- L19:2589 weakly suppresses
- L19:3055 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
- and
- 和
- and
- 和技术
- 和社会
- 性和
- và
- и
- 和完善
- 和个人
Suppresses
- igor
- 溟
- obil
- ceph
- migration
- тем
- emy
- ionales
- еш
- minate