'than' after comparative adjectives
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. 16 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 the word 'than' when it follows a comparison, such as 'happier than' or 'higher than'. On the negative side it fires on 'to' or 'about' after verbs and phrases like 'when it comes to'. Turning it up produced repetitive, looping text.
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.
-
ers Fighting Drowsiness⏎Conclusion. The Employers' Dreams⏎One thing is absolutely certain in America: the quality and quantity of sleep obtained is substantially less than the quality and quantity that are needed. Over the past century, we have reduced
-
to take his perspective.⏎What does this mean for the kiddos of authoritative parents? They grow up knowing how to make their own decisions, and are generally happier than kids who have parents that choose other styles. Does this mean that 10
-
," explains Wrigley.⏎Infants are by far the most vulnerable children in care. Their fatality rate from both accidents and violence is nearly seven times higher than that of children from one to four. Equally striking are differences in infant fat
-
others increase their happiness, it also increases our own. Studies amongst different cultures and ages have consistently found that individuals who spend money on others report higher levels of happiness than those who spend money on themselves. Researchers suggest that the reward we experience from helping
-
cut the emission of carbon dioxide, methane, and other greenhouse gases.⏎It is clearer now than ever that the U.S. and other developed nations can and must act quickly and
-
.S. counties in which new Wal-Mart stores were built between 1987 and 1998 experienced a significantly smaller reduction in their poverty rates than those counties that did not add new Wal-Mart stores.⏎Overall, the portion 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 2000s, 2010s, 2000s, 2000s, 2000s, 2000s, 20
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
3” below the surface) to deep bands at 6” or lower. Deep applications tend to be most beneficial in dry soil conditions. When it comes to timing, Vyn recommends banding P and K in the fall and spring applications
-
, I'm tempted to butt in and ask if you're talking about the Mars rover Curiosity. She's really there! On Mars! Right now
-
are obligated for few formalities. They need to ensure that the professional details, whether relating to the employee or the employer, are to<|endoftext|>It takes a while before one can
-
is, and how little equipment is needed. I’m not that smart; if I can do it, so can you!⏎To be clear, I’m talking about processing (developing, not printing) black and white (35mm or
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
- L10:272 suppresses
- L15:517 excites
- L14:630 weakly suppresses
- L3:2116 weakly excites
- L12:1925 weakly excites
- L10:2035 weakly suppresses
- L4:462 weakly excites
- L5:2671 weakly excites
- L15:2995 weakly suppresses
- L6:1293 weakly excites
- L6:1880 weakly suppresses "by" introducing a deadline or means
- L9:792 weakly suppresses
- L10:155 weakly suppresses
- L10:2098 weakly suppresses
- L14:2150 weakly suppresses
- L6:1787 weakly excites
- L6:914 weakly suppresses
- L1:469 weakly excites
- L3:1292 weakly suppresses
- L4:3054 weakly excites
- L13:2166 weakly suppresses
- L7:138 weakly excites 'to' after verbs like prove or time
- L10:132 weakly excites
- L8:45 weakly suppresses
Feeds later layers
- L17:2192 excites
- L17:865 suppresses
- L19:2397 suppresses
- L18:606 suppresses
- L21:225 excites
- L18:758 suppresses
- L17:1592 excites
- L18:134 excites
- L20:271 excites
- L18:370 suppresses
- L18:2093 excites
- L24:150 weakly suppresses
- L19:1434 weakly suppresses
- L20:482 weakly suppresses
- L18:188 weakly suppresses
- L23:24 weakly excites
- L24:297 weakly excites
- L17:2877 weakly excites
- L18:2271 weakly suppresses
- L19:829 weakly suppresses
- L19:384 weakly suppresses
- L26:731 weakly excites
- L18:2410 weakly excites
- L19:3054 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
- dú
- )[:
- рад
- ),(
- 痕
- /sbin
- 涌现出
- ((⏎
- 涉
- وى
Suppresses
- anship
- elder
- elevation
- Lesser
- Python
- snapping
- .scala
- lies
- !_
- Moose