opening parenthesis after a phrase
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. 25 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 an opening bracket '(' that comes right after a word or phrase, usually to add an explanation. Turned up, the model produced a long string of the letter 'i'. Turned down, it wrote ordinary sentences.
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.
-
“REFUTABILITY”.⏎While scientists are formulating their respective hypotheses, the only way to be objective is to to find out if the hypothesis is falsifiable (
-
threaded gauge will inspect a threaded part based mainly on the pitch diameter. As long as a GO gauge is answered (threads properly without being forced) and a NO-GO gauge is not answered,
-
-level logic circuit that implements a sum of products as shown in Figure 5.11 (a). We may redraw any two-level circuit using a regular structure ( Figure 5.11 b): a vector of buffers, followed by a
-
current.⏎Semiconductors are the individual "on-off" switches that make up a electronic devices. By only allowing the conductance of electricity in specific situations (as opposed to conductors and insulators), semiconductors can produce complex
-
term Kabinett was officially noted in German wine law, and it was given its current definition which applies to wines which are light and non-chaptalized – chaptalization is the process of adding sugar to unfermented grape must
-
protection.It’s simple: the higher the number, the more protection you get. UPF50, for example, means that only 1/50 (2%) of UV Rays will pass through the fabric. As a general rule 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: what is the relationship between the origin of the origin and the origin of the origin? Also, what is the origin of the origin, and what is the origin of the origin origin? We have already learned that the origin of the origin is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
almonds instead of unhealthy foods and lose weight by eating the tasty treat.⏎Another of the health benefits of almonds are good for brain development. The content of tryptoph
-
versal of dysbiosis as a therapeutic model in MS⏎- Roles for short chain fatty acids, bile acids, phytoestrogens, tryptophan, choline and mucin degradation⏎And for those readers interested in data
-
minbar, except in the case of istisqaa’ (praying for rain). And the same applies when making du’aa’ when doing tawaaf or sa’ee.⏎When there is evidence to show that it is permissible
-
's syndrome, the use of haloperidol is limited to the most serious cases.⏎Other medications in this group such as olanzapine (Zyprexa), risperidone (Risperdal), ziprasidone (
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
- L19:158 suppresses
- L23:2010 excites
- L19:1036 suppresses
- L22:978 suppresses
- L21:2772 excites
- L22:144 suppresses
- L20:54 suppresses
- L17:118 excites comma after a clause ending
- L18:2795 suppresses
- L24:35 suppresses
- L20:541 weakly suppresses
- L14:35 weakly suppresses
- L19:400 weakly suppresses
- L18:119 weakly excites
- L18:178 weakly suppresses
- L24:985 weakly suppresses
- L15:65 weakly suppresses
- L17:473 weakly excites
- L21:583 weakly suppresses
- L10:39 weakly excites
- L6:16 weakly excites
- L16:1112 weakly suppresses
- L19:547 weakly excites
- L8:2539 weakly suppresses
Feeds later layers
- L27:719 strongly excites
- L27:878 strongly excites
- L26:820 strongly excites
- L27:996 excites
- L27:1000 suppresses
- L26:1854 excites
- L27:2982 weakly excites
- L26:1041 weakly excites
- L27:509 weakly excites
- L26:416 weakly excites
- L26:1989 weakly suppresses
- L26:1147 weakly excites
- L27:1403 weakly excites
- L26:1577 weakly excites
- L27:2425 weakly suppresses
- L26:2011 weakly suppresses
- L26:1728 weakly excites
- L27:1582 weakly excites
- L26:2626 weakly excites
- L27:1987 weakly excites
- L26:948 weakly suppresses
- L26:581 weakly excites
- L27:2871 weakly excites
- L27:2927 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
- i
- i
- *i
- →i
- {i
- ,i
- )i
- =i
- (i
- .i
Suppresses
- riba
- avras
- unter
- atrás
- ético
- uebas
- ayers
- aine
- auses
- éc