the word 'also' before a verb
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. 11 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
It fires on the word "also" when it sits between a subject and its verb, as in "which also focuses" or "may also be". Turning it up made the model use "also" more often in its replies, like "the sun was also shining".
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.
-
with fireplaces and chimneys.⏎Appropriately and reliably, this issue is also integral to NFPA 211-2019, which also focuses on the reduction of fire hazards associated with the construction and installation of these appliances
-
to hospital for IV therapy, there may not be enough staff to monitor the drip, and the child may die from receiving the wrong amount of fluid, which is also potentially fatal in cases of over-hydration. In order to help health care
-
a multiagent system. In this multiagent system, intersections use a new reservation-based approach built around a detailed communication protocol, which we also present. We demonstrate in simulation that our new mechanism has the potential to significantly out
-
lower mantle. This may change our understanding of the deep Earth.⏎"We still don't fully understand the chemistry of the H-phase," said lead author Li Zhang, also of Carnegie. "But this finding indicates that all geodynamic models need to be
-
span for adult mosquitoes is between 2 weeks to a month or more.⏎Their flight range is usually 50-100 meters. Ranges have also been expressed as 100 feet to 100 yards, and
-
technology to include software and multimedia options that encourage organizations to expand over greater geographic distances has also led to a rise in the use of virtual teams. Virtual teams may also be traditional, self-direct or leadership teams. One challenge of
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 a problem with the first of the first, the first of the first, and the first of the first, and the first of the first, and the first... and the first..." and the first sentence is the first,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, it becomes our problem too. The Aqua Dental clinic at Pearland assure you to be partners in your child
-
special character, three letters⏎- Take a word and insert numbers and special characters in specific positions⏎Those are the basic components of your new password system. Even the simplest version is a fairly difficult password to crack. If you’re really security conscious
-
stood Betsy is a better defense against the modern school system than John Taylor Gatto’s books.⏎“An older girl was reflected (in the mirror) also, near Molly, a dark-eyed, red-cheeked, sturdy little girl,
-
they have refused to host a public hearing. Are they afraid that they would again have a standing-room-only crowd of concerned citizens rejecting the HCP, as they
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
- L4:1458 strongly suppresses
- L7:201 excites
- L10:696 suppresses
- L6:1347 suppresses
- L9:78 suppresses
- L8:845 excites 'too' meaning 'also' at clause end
- L9:598 weakly suppresses
- L5:952 weakly excites
- L8:7 weakly excites
- L5:767 weakly suppresses
- L8:142 weakly excites
- L5:628 weakly excites
- L8:10 weakly excites
- L5:295 weakly suppresses
- L5:872 weakly suppresses 'or' introducing an alternate name
- L5:540 weakly excites
- L8:607 weakly excites
- L2:942 weakly suppresses
- L5:158 weakly suppresses
- L6:7 weakly suppresses
- L10:1676 weakly excites
- L1:2058 weakly suppresses
- L3:172 weakly excites
- L7:1298 weakly excites
Feeds later layers
- L12:2970 excites
- L16:2179 excites
- L13:1302 excites
- L17:2158 excites
- L15:2093 excites
- L17:2081 weakly excites
- L17:2623 weakly excites
- L13:772 weakly excites
- L14:1756 weakly excites
- L13:1533 weakly excites
- L14:858 weakly excites
- L15:1340 weakly excites
- L14:900 weakly excites
- L13:2103 weakly excites
- L13:962 weakly excites
- L16:1688 weakly excites
- L13:995 weakly suppresses
- L15:1693 weakly excites
- L17:962 weakly excites
- L16:1068 weakly suppresses
- L12:2957 weakly excites
- L15:1948 weakly excites
- L12:436 weakly excites
- L19:406 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
- �
- rand
- ます
- рит
- 竹
- ;r
- 出
- rend
- 指
- ately
Suppresses
- 却被
- ignet
- ign
- vig
- 专员
- Boots
- 只见
- kie
- 的伟大
- 奖学金