unclear: conjunctions after a clause
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. 16 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 small connecting words like 'as long as', 'although', 'however' and 'unless the', right where a sentence is about to add a qualifier or contrast. When turned up, the output drifted into rambling, looping text with no clear theme.
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.
-
, physicists have been on the hunt for gravitational waves, ripples in the fabric of spacetime. That hunt is now over. Gravitational waves exist, and we’ve found them.⏎That’s according to researchers at the Laser Interferometer
-
a conversation” or a “description of a feature”. (A BDD story can just as easily describe a non-functional requirement, as long as the work can be scoped, estimated and agreed on.)⏎The structure of a story
-
of physical contact, while 40% reported experiencing verbal abuse.⏎The problem is not limited to female students, however. Male students also experience sexual harassment, although to a lesser extent. According to the same study, 20% of
-
can be made with just about any verb: to do, to be, to go, to make, to take, to see, to get–as long as you use the simple form of the verb. Infinitives can’t be used
-
children are influenced by the behavior of adults, use of electronics by adults is discouraged at Cub Scout meetings and events unless the activity requires the use of them or unless the adult is taking pictures of the event.⏎We understand that many adults have jobs that
-
you). As an Amazon Associate I earn from qualifying purchases. Please read my disclaimer for additional details.⏎Can goats eat strawberries?⏎Yes, goats can eat strawberries, however, not all goats are fond of their sour and sweet taste. Strawberries consist
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 128.128.152.155? I know my computer...? What is the...? I need something? And what is the 128...?
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
with relevant visual imagery.⏎3 Steps to Creating an Awesome Virtual Museum in Class You're spending an afternoon browsing the exhibits at an art museum. If you're anything like me, you'd probably appreciate the art a lot more if you could bring someone
-
evolve as musicians and songwriters and subsequently engineer a truly global revolution in an exhilarating short period<|endoftext|>It's natural for your brain to age over time, just like the rest of your body. However, your diet and lifestyle can also impact the
-
(GNBDK)⏎Never do to anyone else anything that you would not want someone to do to you. “Do not drink so much wine that you get
-
the famous and prestigious Eleven Towns Race that takes place in his country. For years Piet has dreamed of proving that he is a skater as brave and strong as Pim Mulier -- but he had never imagined that his test would fall under
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
- L14:1891 excites
- L15:1702 excites
- L15:1068 suppresses
- L14:531 suppresses
- L14:2150 excites
- L14:655 excites
- L12:896 suppresses
- L6:2699 excites
- L11:3015 suppresses
- L6:502 suppresses
- L12:2215 excites 'of' after 'one' or 'some' (partitives)
- L13:2401 suppresses
- L14:2861 excites 'of' after a percentage
- L14:1128 excites
- L12:1653 suppresses
- L14:778 excites
- L15:1226 suppresses
- L12:2440 suppresses
- L1:1369 suppresses
- L15:1310 excites
- L7:751 weakly suppresses
- L14:1340 weakly excites
- L14:2739 weakly excites
- L15:931 weakly excites
Feeds later layers
- L17:2749 suppresses
- L17:865 weakly suppresses
- L17:57 weakly suppresses
- L17:441 weakly excites
- L24:2586 weakly suppresses
- L19:2671 weakly excites
- L18:773 weakly excites
- L17:2352 weakly suppresses
- L18:2997 weakly suppresses
- L17:201 weakly suppresses
- L17:458 weakly suppresses
- L19:1024 weakly suppresses
- L19:1617 weakly suppresses
- L24:2831 weakly suppresses
- L19:1563 weakly suppresses
- L17:1058 weakly suppresses
- L19:605 weakly suppresses
- L23:238 weakly suppresses
- L19:1778 weakly excites
- L17:775 weakly suppresses
- L19:669 weakly suppresses
- L20:1977 weakly suppresses
- L18:552 weakly excites
- L17:862 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
- pics
- HttpServletRequest
- quisitions
- 刑
- pert
- \Column
- ;]/
- _CID
- RL
- paj
Suppresses
- 儿子
- odon
- ankan
- nty
- arter
- Atom
- adget
- plitude
- side
- sons