words for stretching out or extending
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. 8 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 words that describe something reaching out across a distance, such as 'extends to', 'stretching to the edge', 'juts out' or 'stretches away'. Turning it up did not produce clearly related writing in the test prompts.
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.
-
of the ground must be given.⏎From the lofty plateau of the Chersonese, on which the British army was posted, a long ridge of elevated ground extends to the eastward, on the top of which runs the Woronzoff road.
-
standing up to the test, but Henry III strengthened the defences further. The outer curtain wall was completed, stretching to the edge of the cliff, and Constable's Gate was constructed - a cluster
-
a first floor and a second floor spanning two stories. The outer curtain wall, with twenty individual towers, creates an outer bailey stretching to the edge of the cliff.⏎The Great Tower and forebuilding, viewed from the
-
no park to rest the eyes. The Kennicott family home is outdated and stuffy. The prairie, vast and empty, stretches away on every side.⏎Dreams end and realism begins when Carol takes a thirty-minute
-
of a doubt that our galaxy has a large bar structure that extends halfway out to the sun’s orbit. We know more about where the Milky Way
-
er) to position the⏎arms near the beard, leaving them sticking out in a slightly 3d manner. You⏎can have them sticking straight down, reaching forward, waving, etc
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: I am a 2020,21,22,23 (L) (M) system. Okay, this is the system, and I am just a user. Okay, this is the system, and I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
have a recent version of gcc. m68k-palmos-coff-gcc -v should report its version number to be
-
switched communication, which (once circuit is established) allows immediate start of transfer with a guaranteed bit rate. Confusingly, a communication is often synchronous at the physical layer, while being asynchronous at the data link layer.⏎In programming, asynchronous events
-
place of WEP pending the availability of the full IEEE 802.11i standard. WPA could be implemented through firmware upgrades on wireless network interface cards designed for WEP that began shipping as far back as 199
-
This is quite simple. First, compile (but don't link) each of your .c files, with command lines like the following:⏎m68k-palmos-coff-gcc -g -O2 -c file.c
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
- L6:978 suppresses
- L6:2836 excites
- L7:2307 weakly suppresses
- L7:2802 weakly suppresses
- L6:1730 weakly suppresses
- L4:2989 weakly excites
- L5:60 weakly excites
- L3:862 weakly suppresses
- L5:198 weakly excites unclear: weak "and" in lists
- L2:2530 weakly suppresses
- L6:1010 weakly suppresses
- L5:62 weakly excites
- L7:2933 weakly suppresses "from" after hide, shield, or safe
- L7:2654 weakly excites
- L6:214 weakly excites
- L6:1494 weakly suppresses
- L2:811 weakly suppresses
- L7:278 weakly excites
- L3:2750 weakly excites
- L7:1040 weakly suppresses
- L7:1791 weakly suppresses
- L6:2490 weakly excites
- L7:1608 weakly excites
- L7:2792 weakly excites
Feeds later layers
- L10:2980 suppresses
- L11:1039 suppresses
- L13:2276 excites
- L10:1400 suppresses
- L16:2602 excites
- L9:438 weakly suppresses
- L19:3008 weakly excites 'the' or 'outside' after 'cut off from'
- L22:3014 weakly excites
- L11:3029 weakly suppresses
- L13:2851 weakly excites
- L16:848 weakly suppresses
- L10:2982 weakly suppresses unclear: product and model names
- L19:2581 weakly suppresses
- L22:2984 weakly excites
- L21:79 weakly suppresses
- L19:956 weakly excites
- L12:299 weakly excites
- L11:1073 weakly suppresses
- L17:2446 weakly excites
- L22:2371 weakly excites
- L10:2430 weakly excites
- L27:661 weakly suppresses
- L9:607 weakly excites
- L13:1872 weakly excites comparative or contrasting qualifiers
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
- esis
- ilate
- inction
- tận
- orna
- pheres
- tte
- InParameter
- tics
- émon
Suppresses
- type
- 是一种
- -base
- lectic
- 存在的
- 三种
- 类型
- 一种
- 来源
- ooth