first piece of a split name
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. 27 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 first part of a name or word that is broken into several pieces, like the 'R' in Rizal or the 'Cho' in Cho Oyu. Turned up, the model fell into repeating phrases like 'experienced experienced'.
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.
-
and with the exception of sweet corn and peas, all are subject to chilling injury. Fruit vegetables are not generally adaptable to long-term storage. Exceptions are the hard rind (winter) squashes and pumpkin. A useful classification for postharvest
-
Lap of the Mighty Sagarmatha - Solu Khumbu or Everest region⏎The major mountains are the Mt.Everest, Mt.Lhotse, Cho Oyu, Nuptse, Pumori, Ama Dablam
-
that fell under the class leadership of the ilustrados or the liberal bourgeoisie⏎that surrounded Aguinaldo.⏎In this Propaganda Movement, Dr. Jose Rizal made patriotic annotations on Morga<|endoftext|>Early in the 14th
-
someone is stuck in the gap.⏎- Uses existing camera infrastructure.⏎- Processes in real time, with low latency on a computer the size of a credit card (Raspberry Pi for example.)⏎- Improved safety at the PTI⏎- Optimisation
-
coconut water)⏎– kombucha (fermented tea)⏎– rejuvelac (fermented sprouted grains)⏎– kvass or kvas (fermented rye drink)⏎Define fermented foods. What foods are fermented?⏎In order to ferment
-
the US state of Alaska lies the 20,320 ft (6,194 m) Mount McKinley or Denali (its Koyukon Athabaskan name meaning “The Great One”). The majestic mountain,
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: while at work at nightfalls into some distastefully distastefully distastefully distastefully distastefully distastefully distastefully distastely distastely distastely distastely distastely distastely dist
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to environmentally relevant behaviors. This observation may inspire researchers to⏎more directly test predictions arising from current theoretical models,⏎gather data on actual behavior rather than relying on self-reports,
-
proportion.⏎5. (n.) To exercise the rational faculty; to deduce inferences from premises; to perform the process of deduction or of induction; to ratiocinate; to reach conclusions
-
gents, fertilizers, construction and oil&gas materials. The documents are available at this link (https://echa.europa.eu/it/registry-of-restriction-intentions/-/dislist/details/0b0
-
it works faster and more efficiently to arrive at the truth. To find and avoid science’s seductive pitfalls, researchers are studying science itself, a field called meta-research.⏎Now, a paper by three Stanford researchers that delves into those pitfalls
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
- L23:1368 excites
- L22:961 suppresses
- L17:253 suppresses
- L2:19 suppresses
- L22:2630 suppresses
- L20:257 suppresses
- L17:420 suppresses
- L23:206 suppresses
- L18:343 suppresses
- L20:97 suppresses
- L20:669 suppresses
- L24:105 suppresses
- L21:475 weakly excites
- L10:273 weakly suppresses
- L18:151 weakly suppresses
- L16:17 weakly suppresses
- L7:300 weakly excites
- L20:52 weakly suppresses
- L18:1587 weakly suppresses
- L20:151 weakly excites
- L26:495 weakly suppresses
- L2:1762 weakly suppresses
- L17:93 weakly suppresses
- L0:1365 weakly suppresses
Feeds later layers
None: this is the last layer.
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
- lore
- SessionFactory
- 你会发现
- TRL
- знач
- SOURCE
- стоя
- enlisted
- .cl
- Src
Suppresses
- falls
- fer
- ῃ
- 日在
- sthrough
- odom
- 之夜
- _ne
- erto
- 市中心