"in" or "real" after 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. 6 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 small connecting words like "in" right after an action word ("fly in tighter flocks", "operating in high temperatures") and on "real" in "real time". When it is turned up, the model's writing drifts in unrelated directions, so the effect is hard to see.
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 deeper wingbeats and proportionately narrower and less fingered wings. They also have a shorter, thicker neck, a much shorter bill and frequently fly in tighter flocks. The underwing is uniformly grey, unlike choughs.
-
turbomachinery is limited when operating in high temperature environments, which restricts the efficiency of gas turbines and internal combustion engines
-
line.)⏎- Sends alerts to drivers, staff and other stakeholders if 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 (R
-
cannot provide full torque continuously when the rotor is stopped. This aids in maintaining the current position.⏎- Precise open-loop positioning and repetition. Stepper motors move in discrete steps as long as the motor stays under the maximum torque and current limits.
-
sense of responsibility towards them.⏎The companions<|endoftext|>The Internet of Things (IoT) is about connecting any device to the internet and enabling it to transmit data in real time without human intervention. This technology makes it possible for users to control their smart
-
"tuneful, beatful, and artful." This is someone who: is able to sing in tune as well as catch tunes and is comfortable singing alone and with others; is
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 person who I hold up to, the man who I respect for, is not only the great, but also the person who I have been able to be. I have admired, and I have learned from, and I have respected,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
were real people" (11). He continues "it's plausible but unproven" and that is good enough for me.⏎Why do I love and admire Achilles so much? Let me count the ways. The "Greatest of the
-
all great heroes of action was the Greek Hercules, who might almost be regarded as idolised toil. The adoration of this semi-god actually exceeded the reverence accorded to many of the twelve gods of the Greek and Roman Olympus. This
-
acclaimed legend in her lifetime, a towering legacy left behind. Rest in peace, brave comrade,” she penned.⏎Former
-
God is the only thing worthy of worship. Any other gods that others believe in, then we would regard them as false
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
- L2:1148 weakly suppresses
- L3:258 weakly suppresses
- L2:2032 weakly suppresses
- L5:2682 weakly excites
- L3:93 weakly suppresses
- L0:2762 weakly excites
- L4:1178 weakly excites
- L5:1936 weakly suppresses
- L4:1631 weakly suppresses
- L2:2250 weakly suppresses
- L0:2680 weakly suppresses
- L4:1059 weakly excites
- L4:1912 weakly excites
- L1:747 weakly suppresses
- L5:2334 weakly suppresses
- L0:2839 weakly suppresses
- L4:2881 weakly suppresses
- L2:2641 weakly excites
- L5:3010 weakly excites
- L4:1637 weakly suppresses
- L2:1166 weakly suppresses
- L2:1073 weakly suppresses
- L3:1017 weakly suppresses
- L4:2105 weakly suppresses
Feeds later layers
- L16:2658 weakly excites
- L7:2701 weakly excites
- L12:1601 weakly suppresses
- L7:1526 weakly excites
- L8:2225 weakly suppresses
- L9:2571 weakly suppresses
- L12:2764 weakly excites
- L17:1874 weakly excites
- L25:2219 weakly excites
- L23:2595 weakly suppresses
- L9:2723 weakly excites
- L18:1837 weakly excites
- L18:710 weakly suppresses
- L17:2395 weakly excites
- L25:635 weakly suppresses
- L22:574 weakly excites
- L18:2034 weakly excites
- L21:2482 weakly suppresses
- L19:1321 weakly suppresses
- L8:2661 weakly excites
- L18:2541 weakly suppresses
- L24:1233 weakly excites
- L13:2997 weakly excites
- L10:294 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
- WXYZ
- OOK
- iday
- unny
- rary
- IJ
- ijk
- óng
- 衽
- proof
Suppresses
- امة
- ๆ
- 殿堂
- че
- foremost
- ellar
- Installed
- SqlCommand
- philosoph
- utzer