unclear: word closing a two-word term
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. 2 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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
It fires on the last word of a short phrase or name, like 'circuit' in 'short-circuit', 'Pacific' in 'Great Pacific' or 'track' in 'fast track'. The pattern is loose and the steering outputs don't show a clear theme.
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.
-
you remember when you learned multiplication and discovered multiplication was a faster and shorter way to represent repeated addition? Well there is a faster and shorter way to represent repeated multiplication.⏎Fractions - Adding Mixed Numbers⏎Learn how to Add
-
C. Marshall, who was the US Secretary of State. Thousands of millions of dollars were provided for rebuilding cities, roads, industries etc⏎Definition from the Longman Dictionary of Contemporary English Advanced Learner's Dictionary.⏎Dictionary pictures of the day⏎
-
of fault lines circling the Pacific Basin that is prone to frequent earthquakes and volcanic eruptions.⏎CNN's Kathy Quiano contributed to this report.<|endoftext|>From Longman Dictionary of Contemporary Englishldoce_734_znoticeno‧t
-
Maximum short-circuit current⏎18. Current acquisition channel⏎19. The internal resistance of
-
or short-circuiting. A fuse works by breaking an electrical circuit if the current flowing through the
-
to ensure accuracy in their projects. Similarly, many new software development projects keep engineering applications in mind during their development. Young engineers looking to put themselves on the fast track to success would benefit greatly from learning computer technology and science applications relevant to their field
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: " The following is the problem: " A man's in the first place in the second place in the third place in the fourth place in the fifth place, and the sixth place, and the seventh place, and the eighth place, and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
or by contacts with immunogens.⏎||Conferred by introduction of ready-made antibodies.⏎||Immune response is durable and effective.⏎||Immune response is short lived and less effective.⏎||Immunity develops only after a lag period.⏎||Imm
-
to staying up-to-date, boosting your immune system is wellness 101. Better yet, it’s simpler than you might think (and tastes great, too!). Additionally
-
, he became infected three years ago from using dirty needles. Thanks to a daily cocktail of 12 anti-retroviral drugs which stops him developing full-blown AIDS, Vladimir can now live an almost normal life. Today, he spends his
-
can disrupt the pathway by binding to certain factors and preventing them from cleaving and activating other factors. In this way, coagulation is circumvented.⏎Sure enough, Simukunin binds to clotting factors and interferes with coag
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
- L0:1075 excites
- L1:861 suppresses
- L0:191 weakly suppresses
- L0:2641 weakly suppresses
- L1:491 weakly excites
- L1:2677 weakly suppresses
- L1:1029 weakly excites
- L1:647 weakly excites
- L0:1301 weakly suppresses
- L0:903 weakly excites
- L0:178 weakly suppresses
- L1:1874 weakly excites
- L1:89 weakly excites
- L0:2415 weakly suppresses
- L0:280 weakly suppresses
- L0:1224 weakly suppresses
- L1:1340 weakly suppresses
- L1:257 weakly suppresses
- L1:2291 weakly suppresses
- L0:1814 weakly suppresses
- L1:2164 weakly suppresses
- L1:791 weakly suppresses
- L0:2842 weakly suppresses
- L1:401 weakly suppresses
Feeds later layers
- L4:656 suppresses
- L11:922 weakly suppresses
- L5:2885 weakly suppresses
- L24:1059 weakly excites
- L3:844 weakly suppresses
- L14:411 weakly excites
- L4:723 weakly suppresses
- L9:1906 weakly suppresses
- L9:1834 weakly suppresses
- L13:200 weakly excites
- L8:1012 weakly suppresses
- L13:2289 weakly excites
- L8:825 weakly suppresses
- L16:1148 weakly excites
- L13:2898 weakly excites
- L23:1152 weakly excites
- L4:3058 weakly suppresses
- L4:1441 weakly suppresses
- L11:2994 weakly suppresses
- L8:525 weakly suppresses
- L7:171 weakly suppresses
- L27:979 weakly suppresses
- L12:849 weakly excites
- L13:1819 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
- 贯
- contenu
- jer
- казалось
- 呗
- 钢铁
- сет
- 煤炭
- 权威
- 극
Suppresses
- ritch
- rowser
- ancy
- ساء
- .gc
- ortal
- وغير
- .iOS
- ��
- doInBackground