"the" in mid-sentence, weak
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. 20 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 weakly responds to the word "the" in the middle of a sentence, especially after a verb or preposition. Its strongest opposite reactions come on words like "over", "and" and "than" in comparisons such as "one over the other". The steered text was loosely about lists and comparisons, but not clearly.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
you need to ask before you can identify an individual?⏎When you’re looking at twenty-four candidates, the worst case scenario is twenty-two (assuming it’s not the person on your card) if you just ask “Is your person [name]
-
otherwise, the two will ot stick together. If it is the job of a protein to catalyze a chemical reaction, then the shape of
-
second bottle to be the fake.⏎In practice, there are some limitations to a manual application of this geometric
-
: Associated Press⏎About one in seven pregnant women get flu shots even though health experts encourage them to be the first in line every flu season.⏎Many obstetricians don't vaccinate and
-
stocks that have dividend yields of 15% or more. To a beginner investor, stocks like this may seem like a bargain. However, it is often the case that the dividends for these high yielding companies are unsustainable and only high because the
-
stomach cancer than white adults.⏎- In the poorest U.S. counties, men with colorectal cancer have a 35% higher death rate than men in the most affluent counties.⏎- Bisexual women are 70% more likely to
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 fact that the word the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ivalent’ non-PWM fan, but non-PWM fans can still run slow enough. For the average consumer I can’t think of much reason to prefer one over the other.⏎Can you control RGB colors
-
pace. To switch between modelspace and paperspace, view the tabs located at the bottom of the screen. One tab is labeled ‘modelspace’ and the other tabs are labeled either as ‘sheet’ or ‘layout’. The ‘
-
both good options when it comes to charging your battery efficiently. When differentiating PWM and MPPT, the discussion is not going to be about which one is better than the other.⏎What should be determined when choosing between PWM and MP
-
.⏎Counseling Point for Anti-Platelets:⏎Aspirin and warfarin are sometimes prescribed together, and this is because one is anti platelet and the other is anticoagulant. Please see the previous anticoagulant
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
- L19:2606 strongly excites
- L19:1933 strongly excites
- L19:2013 excites
- L18:2251 suppresses
- L18:2780 suppresses
- L17:2968 excites
- L18:2605 suppresses
- L19:2745 excites
- L19:2789 excites
- L18:2915 suppresses
- L19:3028 suppresses
- L17:2957 excites
- L17:1724 suppresses
- L18:2953 excites
- L19:2283 suppresses
- L19:2418 excites
- L19:2587 excites
- L18:1458 suppresses
- L16:1510 weakly excites
- L4:854 weakly suppresses
- L18:2241 weakly excites
- L19:2703 weakly excites
- L18:2256 weakly excites
- L15:130 weakly suppresses
Feeds later layers
- L26:507 strongly suppresses
- L27:127 strongly excites
- L21:863 strongly suppresses
- L27:399 strongly suppresses
- L23:94 excites
- L24:49 excites
- L21:758 suppresses
- L26:535 excites
- L22:199 suppresses
- L22:171 suppresses
- L27:1464 excites
- L23:371 excites
- L25:2531 suppresses
- L27:1382 weakly suppresses
- L26:1200 weakly suppresses
- L27:2537 weakly excites
- L27:994 weakly excites
- L25:2497 weakly suppresses
- L21:1232 weakly suppresses
- L27:121 weakly excites
- L25:1796 weakly suppresses
- L21:488 weakly suppresses
- L27:25 weakly suppresses
- L26:2278 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
- others
- 除此之外
- Others
- 其他
- others
- OTHER
- another
- Others
- another
- other
Suppresses
- the
- the
- The
- _the
- The
- THE
- -the
- →the
- →The
- ethe