unclear: pieces of split words
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. 1 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 middle pieces of words that have been chopped into fragments, most strongly in 'laissez-faire' and in surnames like 'Dupriez'. Turning it up made the model's output drift into loose, unrelated sentences, so it does not clearly steer any topic.
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.
-
Pine Ridge, he advocated for peace with the European Americans⏎Date de dernière mise à jour : 05/07/2021⏎Ajouter un commentaire<|endoftext|>By DANIEL SUDDEATH⏎NEW ALBANY —
-
of placing the retaliatory use of force under objective control.⏎When I say “capitalism,” I mean a full, pure, uncontrolled, unregulated laissez-faire capitalism — with a separation of state and economics, in the same way
-
.”12⏎However, banning research is also unlikely to be desirable as rogue research could still continue.2 This may lead to a worse outcome than a laissez-faire approach because clandestine science has the capacity to drive cold wars for decades
-
(at least some of them).⏎Eventually, Golway argues, Tammany Hall introduced the notion that the government had a social welfare role to play. Laissez-faire capitalism needed some help, and the corrupt politicians of that era helped push
-
3(3), pages 315-320.⏎- Angus Deaton & Olivier Dupriez, 2011.⏎"Purchasing Power Parity Exchange Rates for
-
.⏎(2) Internal data – Laboratoires URGO – QIPI VLU 2019 / 4⏎(3) Améliorer la qualité du système de santé et maîtriser les dépenses : propositions
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 been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
BPA. Glass bottles, however, also leach chemicals, including aluminum, explains a study by Clemens Reimann of the Norges Geologiske Undersøkelse in Norway and colleagues. Aluminum is sometimes added to soda-l
-
3(3), pages 315-320.⏎- Angus Deaton & Olivier Dupriez, 2011.⏎"Purchasing Power Parity Exchange Rates
-
Col. Stand Watie, with one regiment of Cherokees, to move into the neutral land and
-
de Vereenigde Doopsgezinde Gemeente to Amsterdam, 2 vols. Amsterdam: Uitgegeven en ten geschenke aangeboden door den Kerkeraad dier Gemeente, 1
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
- L0:1301 suppresses
- L0:178 suppresses
- L0:1224 suppresses
- L0:1193 weakly excites
- L0:1 weakly suppresses
- L0:894 weakly excites
- L0:489 weakly excites
- L0:1921 weakly excites
- L0:1167 weakly suppresses
- L0:220 weakly suppresses
- L0:75 weakly suppresses
- L0:429 weakly excites
- L0:2641 weakly suppresses
- L0:84 weakly suppresses
- L0:1468 weakly suppresses
- L0:30 weakly suppresses
- L0:1836 weakly suppresses
- L0:2094 weakly excites
- L0:25 weakly excites
- L0:485 weakly excites
- L0:2542 weakly excites
- L0:48 weakly excites
- L0:91 weakly suppresses
- L0:386 weakly suppresses
Feeds later layers
- L2:765 excites
- L2:93 excites
- L3:104 excites
- L3:110 suppresses
- L27:149 weakly suppresses
- L6:819 weakly excites
- L13:612 weakly excites
- L3:970 weakly excites
- L3:2807 weakly excites
- L25:307 weakly excites
- L27:108 weakly suppresses
- L4:266 weakly suppresses
- L23:1780 weakly excites
- L19:1788 weakly excites
- L6:31 weakly excites
- L9:413 weakly suppresses
- L3:74 weakly excites
- L9:584 weakly suppresses
- L4:420 weakly excites
- L4:528 weakly excites
- L7:2490 weakly excites
- L5:928 weakly excites
- L2:26 weakly excites
- L23:1561 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
- =DB
- десь
- babel
- =db
- acular
- =Value
- =""><
- Weber
- 的一款
- parable
Suppresses
- yk
- ULT
- dom
- 人在
- DEAL
- hari
- istically
- ettle
- alties
- -domain