word-start 'mult' fragment
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 piece 'mult' at the start of longer words such as 'multivitamin', 'multitudes' or 'multilayer'. Turned down, the model produced a long run of a Chinese counting word, and turned up it wrote loosely on 'the fact that' and similar phrases.
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.
-
Sickness reliever), Azadirachta indica (Neem), belonging to the Meliaceae family, from prehistoric Indian subcontinent, for its multitudinous medicinal properties (Tanguturi et al. 202
-
''s most intelligent animals. See a great silverback gorilla, the monkey with the biggest nose, how gibbonsswing through the trees, multicolored marmosets, and a day in the life of a gor
-
Words with Power, these words slimply leaped up the page:⏎Then again, Christ is constantly associated with the miraculous provision of food. Miracle of feeding large multitudes with very small amounts of food are recorded in all four Gospels, sometimes
-
C, 150mg magnesium, 1500mg calcium and 500 mg niacin from dietary sources each day. A good multivitamin/mineral supplement (like Centrum) is also recommended.⏎- Omega
-
the effects of high doses of particular vitamins or minerals, like calcium and vitamin D, this clinical trial examined whether a common daily multivitamin had an effect on overall cancer risk. A randomized, double-blinded
-
affordable emissions free car The question an issue of engineering and economics is being studied by ... - Read More⏎Researchers at Rice University have created flexible patterned sheets of multilayer graphene from a cheap polymer by burning it with a computer controlled laser The
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:个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个个
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
81 launched a scheme called Fifth Generation, 10-year program for the development of supercomputers was presented in it.⏎Britain created a project called ‘Elvi’ whereas European Union countries also introduced a program called ‘Esprit’.⏎
-
If exp is 0, x is returned.⏎These functions will fail if:⏎The result overflows.⏎If the integer expression (math_errhandling & MATH_ERREXCEPT) is non-zero, the overflow floating-point exception is raised.⏎
-
of the commoner.⏎This phrase originates from Rome in Satire X of the Roman satirist and poet Juvenal (circa A.D. 100). In context, the Latin metaphor panem et circenses (bread
-
activate subtitles in english for better understanding.⏎John Forbes Nash Jr., a mathematical genius whose struggle with schizophrenia was chronicled in the 2001 movie “A
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
- L23:2883 strongly excites
- L22:2314 excites
- L9:59 suppresses "a" meaning per (a day, a year)
- L4:326 suppresses
- L11:233 suppresses
- L19:1935 excites
- L7:224 suppresses
- L20:311 suppresses
- L17:1818 excites
- L21:708 suppresses
- L4:961 excites
- L4:205 excites the digit 2 inside citations
- L16:600 suppresses
- L18:973 excites
- L21:1863 suppresses
- L14:2220 weakly excites "this" pointing back to a thing
- L4:731 weakly excites
- L18:2629 weakly suppresses
- L18:2804 weakly suppresses
- L24:2344 weakly suppresses
- L9:13 weakly excites
- L19:3060 weakly suppresses
- L18:2183 weakly suppresses
- L18:635 weakly excites
Feeds later layers
- L27:1694 weakly excites
- L27:622 weakly excites
- L27:994 weakly suppresses
- L27:653 weakly suppresses
- L27:2365 weakly suppresses
- L27:1908 weakly excites the 'd' that starts 'die' or 'dead'
- L27:2701 weakly excites
- L27:925 weakly excites
- L27:2744 weakly suppresses
- L27:2743 weakly suppresses
- L27:1522 weakly excites
- L27:1467 weakly suppresses
- L27:3017 weakly suppresses
- L27:2271 weakly excites
- L27:1782 weakly suppresses
- L27:1276 weakly suppresses
- L27:1425 weakly suppresses
- L27:2599 weakly suppresses
- L27:624 weakly suppresses
- L27:193 weakly excites
- L27:2393 weakly suppresses
- L27:703 weakly suppresses
- L27:1544 weakly excites
- L27:469 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
- illance
- ellation
- ishments
- acle
- ulary
- ivals
- /cms
- utt
- ishing
- AMERA
Suppresses
- 个
- 个小
- 种
- 些
- 個
- 个工作
- 个城市
- 个多
- 个国家
- 个项目