unclear: scattered mid-phrase 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. 27 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 a grab bag of words in the middle of sentences, with no clear shared theme. When it was turned down, the model got stuck repeating 'Ob'. Turned up, the output drifted into unrelated rambling.
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.
-
is considered transverse when the fracture extends across the bone.⏎SpiralAs you may have guessed, a spiral fracture spirals around the bone.⏎ObliqueAn oblique fracture runs diagonally across the bone⏎ComminutedA commin
-
colors to identify teams.⏎Shows an obelisk – is a monument that has been installed on the spina. We know that there is one circus that had this Egyptian object installed on it – the Circus Maximus.
-
and when it exceeds the apneic threshold PCO2, breathing resumes.⏎- Obstructive sleep apnea (OSA) is due to upper airway occlusion secondary to neuromuscular failure, primarily involving hypoglossal
-
is known of the<|endoftext|>Obese & Fat German Shepherd⏎Like humans, dogs can become obese for several reasons, and the effect is just as severe in them. German Shepherds are prone to becoming obese, and in this article, you
-
students’ basic competence to solve numeracy problems, the students’ obstacles in finishing numeracy problems, and the students’ strategy to overcome the difficulties.<|endoftext|>Gasoline prices have increased rapidly during the past several years, pushed
-
in upper airway closure (obstructive apnea, Fig 2) or narrowing (obstructive hypopnea). Obesity is the major risk factor for OSA, in part due to excess fat deposition in the pharynx.
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:obobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobobob
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
” system that can recognize individual animals by their unique skin patterns, which are as distinct as a human fingerprint.⏎The park, dotted with muddy water holes and ancient baobab and acacia trees,
-
ürkiye’de doğdu. Lisans öğrenimini MSÜ Deniz Harp Okulunda tamamlamıştır.⏎Yüksek Lisans Öğrenimine MSÜ Barbaros DEBİM Enstitüsünde dev
-
aiming for good mileage or even great mileage, hypermilers seek to push their gas tanks to the limit and achieve hypermileage,
-
, banana, jackfruit, papaya, melons, guava, passion fruit, custard apple and avocado pear are common. Many traditional fruits such as baobab, wild custard apple, carissa, dialium, flac
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
- L23:1115 suppresses
- L23:2134 excites
- L22:1775 excites
- L25:827 suppresses
- L23:632 excites
- L26:2779 excites
- L26:1997 excites
- L22:808 suppresses
- L23:1090 excites
- L17:839 suppresses
- L2:2594 weakly excites
- L25:1228 weakly suppresses
- L24:950 weakly suppresses
- L19:2091 weakly excites
- L11:2626 weakly excites
- L21:1505 weakly excites
- L10:2273 weakly suppresses
- L22:250 weakly suppresses
- L22:2839 weakly excites
- L12:2993 weakly suppresses
- L21:1651 weakly excites
- L17:1621 weakly excites
- L26:1633 weakly excites
- L25:2142 weakly excites
Feeds later layers
None: this is the last layer.
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
- egas
- cka
- apult
- arness
- 桢
- anoia
- annel
- чная
- 镝
- ança
Suppresses
- ob
- Ob
- →ob
- ob
- Ob
- _ob
- (ob
- -ob
- .ob
- OB