'the' in 'in the long run'
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. 26 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 'the' in phrases like 'in the long run' or 'in the end', where 'in the' means 'eventually'. Turning it up made the model get stuck repeating words like 'medium' or 'very', which does not relate to the trigger.
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.
-
of your diet. Although processed food has an undeniable convenience factor and longer shelf life, they typically contain unnecessary additives and chemicals that can be detrimental to your health in the long run.⏎While not always the budget-friendly option, there are many benefits to
-
ay/grain, and finally in extreme cases you might even consider selling productive livestock. In many cases, making the hard decisions early can be the best decision in the long-run.⏎Be ready with nitrogen⏎Pastures and livestock can make compensatory
-
solar energy is environmentally friendly. It does not produce harmful emissions like some other forms of energy production do. The third reason is that solar energy is cost-effective in the long run.⏎Solar mobile power sources are a cost-effective alternative to UPS systems⏎
-
they would be more efficient and the world would be slightly more sustainable in the reduction of energy and time used by humans to create robots for a better world.
-
. This may lead to a difficulty in controlling your metabolism and a general slowing of the metabolism. The latter of these typically makes it much harder to lose weight in the end. Yoyo dieting may also put other strains on different parts of your
-
which is more resilient and cost-effective in the long term. Local governments need systems that enable them to incorporate the health of natural
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: If I really want to understand what reallyiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaia
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in their community. I love the story of Luiz.⏎Lisa Baldwin, a chemistry teacher, works with her students to fight through academic challenges.⏎By MATT RICHTEL⏎There is a widespread
-
content, color, weight, and construction all factor into what level is offered.⏎Based on ASTM International testing, products that feature l<|endoftext|>How Dick Cheney & Donald Rumsfeld Helped Richard Nixon Implement The Worst Idea Ever⏎Conor Frieders
-
The Big Picture? An Afterword⏎PortraitDyan Mazurana is senior research fellow at the Feinstein International Famine Center, Tufts University. Angela Raven-Roberts is director of research and programs at the Feinstein Famine Center, Tu
-
burg and Iredell, and several mines, some still in operation, in Stanly county in the New London/Richfield area. But Gold Hill, just a few miles from Stanly and
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
- L16:107 excites
- L20:185 weakly excites
- L19:572 weakly excites
- L2:2028 weakly suppresses
- L11:2758 weakly suppresses
- L17:166 weakly suppresses
- L18:282 weakly suppresses
- L5:2175 weakly suppresses
- L10:1745 weakly suppresses
- L24:1223 weakly suppresses
- L20:164 weakly suppresses
- L16:305 weakly suppresses
- L20:1713 weakly suppresses
- L19:277 weakly excites
- L2:1866 weakly suppresses
- L21:2096 weakly suppresses
- L16:463 weakly suppresses 'for' introducing a purpose or target
- L16:137 weakly excites
- L17:253 weakly suppresses
- L0:41 weakly suppresses
- L18:137 weakly excites 'the' after 'to' in fixed phrases
- L10:12 weakly suppresses
- L22:316 weakly suppresses
- L10:503 weakly excites
Feeds later layers
- L27:1178 strongly excites
- L27:1745 weakly excites
- L27:2621 weakly suppresses
- L27:842 weakly excites
- L27:2280 weakly excites
- L27:1382 weakly excites
- L27:1221 weakly excites
- L27:2747 weakly suppresses
- L27:1165 weakly excites
- L27:1810 weakly excites
- L27:2240 weakly excites
- L27:2429 weakly excites
- L27:2638 weakly excites
- L27:930 weakly excites
- L27:1089 weakly excites
- L27:2885 weakly suppresses
- L27:1339 weakly suppresses
- L27:837 weakly suppresses
- L27:274 weakly excites document boundary, code keywords next
- L27:333 weakly excites
- L27:3054 weakly excites
- L27:2600 weakly excites
- L27:1733 weakly excites
- L27:1951 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
- long
- long
- Long
- Long
- →long
- 长
- (long
- .long
- -long
- LONG
Suppresses
- 骀
- ricing
- neau
- odox
- CAA
- äter
- 鹃
- ogn
- oge
- IONS