'had' after has or have
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 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 responds to the word "had" when it comes straight after "has", "have" or "haven't", as in "has had a significant impact". Turning it up did not produce writing that clearly shows this pattern.
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.
-
Bronfenbrennerdeveloped the ecological model ofhuman development. Urie Bronfenbrenner8217;s ecological model of human development has had essay junk food should be banned enduring impact ambition essay on macbeth thefield of
-
or they haven’t had the proper training. It’s up to the manager from this point to ensure the
-
generation that has had such a chance for peace, for development, for wealth, for freedom for other
-
-dimensional plastic objects and polymer films. The recent introduction of high-density atmospheric plasma for film, foil, paper, foam, nonwoven and woven substrate applications has had a significant impact on the surfaces of these materials at considerably lower capitalization costs.⏎
-
Natural Capital Accounting, or NCA. Through NCA, natural resources can be mainstreamed into development policy and planning. This video captures the impact that accounting has had on policy-making in Indonesia.⏎Last year, WAVES program stated that “A
-
grains come in many shapes and sizes, from large kernels of popcorn to small quinoa seeds.⏎- Whole grains. These are unrefined grains that haven't had their bran and germ removed
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: been given to a given the following 100% for the given given given given given given given given given given given given given given given given given given given given given given given given given given given given given given given given given given
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
would also affect us as without any doubt.⏎Enormous amount of plastic waste are overflowing
-
historian working within the Classic Catholic Framework (CCF) will face all the diverse and varied expressions of Christian belief brought forth from the relevant texts. He will, however, ask different questions about those texts from those who work in the CPF or the
-
of clothing should also keep the heat away from the body of the person wearing it.⏎Facility Clothing
-
should also meet healthcare standards for washability, mold and microbial resistance, and fire safety.
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
- L0:1730 weakly excites
- L1:1055 weakly excites
- L1:2411 weakly suppresses
- L0:1228 weakly suppresses
- L1:1693 weakly suppresses
- L0:580 weakly suppresses
- L1:139 weakly excites
- L0:377 weakly excites
- L1:135 weakly suppresses
- L1:1662 weakly suppresses
- L1:3000 weakly suppresses
- L0:1541 weakly suppresses
- L1:1321 weakly suppresses
- L1:604 weakly excites
- L0:2474 weakly suppresses
- L0:732 weakly excites
- L1:838 weakly suppresses 'be' after modal verbs like 'can'
- L1:2872 weakly suppresses
- L1:1561 weakly excites
- L0:1301 weakly suppresses
- L0:2587 weakly excites
- L1:491 weakly excites
- L1:2128 weakly excites
- L0:2095 weakly suppresses
Feeds later layers
- L3:309 excites
- L4:113 suppresses
- L3:2198 suppresses
- L7:1406 excites
- L8:403 suppresses
- L4:124 suppresses
- L18:2266 weakly suppresses
- L14:622 weakly suppresses
- L3:1202 weakly excites
- L3:756 weakly excites
- L14:2417 weakly suppresses
- L4:1043 weakly excites
- L9:95 weakly suppresses
- L21:61 weakly suppresses
- L16:1935 weakly excites
- L9:1903 weakly excites
- L20:1338 weakly excites
- L6:2398 weakly excites
- L4:48 weakly suppresses
- L12:124 weakly suppresses
- L10:3016 weakly suppresses
- L7:1580 weakly excites
- L14:898 weakly excites
- L4:2128 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
- электро
- groupName
- orer
- 硬度
- ぶり
- mse
- ,filename
- masa
- filename
- ạp
Suppresses
- vent
- thers
- S
- donn
- ic
- Literal
- ley
- ��
- ************************************************************************
- ox