"as" or "since" giving a reason
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. 9 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 word "as" (or "since") when it starts a clause that gives a reason, like "the test is hard to pass, as...". Turning it up did not produce a clear topic in the writing.
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.
-
, the maximum DSR typically allowed by lenders is around 60%.⏎For those individuals with irregular income, it may be harder to pass the stress test as
-
aka CLR. It do all the heart tasks of locating, loading and managing .NET types. This is a very important task to be done by the CLR, as it guarantees that the written types will function properly according to what
-
9% of the total population. This comes out to roughly 1.8 million Kurds living in Syria. This number, however, is anything but accurate as the Syrian government does not keep figures. Also the definition of ‘Kurd’
-
or as a takeout container from a fast food place⏎- egg cartons – either the styrofoam ones or cardboard ones (latter is preferable as you can plant these in the ground and they will rot away)⏎- plastic containers
-
inct pacing profiles result in similar perceptual responses and neuromuscular fatigue development⏎I found this study quite interesting as I've been investigating different pacing strategies. The authors wanted to understand "the effect
-
can be completely operated via the serial interface with a teletype. By the way, it is not possible to show the internal registers at the lamp registers, as it is possible at a processor built with the TTL technology.⏎Right beside the front
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 use of, or the display of, a screen, or a display device, or a display of a screen for the purposes of communication, or for the purposes of displaying visual information, or for the purposes of transmitting data, or for
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
academic community use digital technologies to preserve the scholarly record and to advance research and teaching in sustainable ways ... Types of Sources for a Research Paper | Pen and the Pad There are two types of research<|endoftext|>The Apollo 11 space flight landed the first
-
, and a conclusion.⏎It is important that the consistency of the format be precise using the same paragraph spacing and font style/size throughout content of report. AgendaAn agenda informs the attendees, of what is going to happen during the meeting and help
-
concerns relating to privacy, ethics, and legality. The government has taken a recent active role in regulating online activity. Exploring Our Inner-self in Cyberspace Cyberspace is a new communication medium which enables us to understand our social behavior.
-
to understand how current works.⏎Current is the rate at which electrons flow past a point in a complete electrical circuit.⏎Want to Slash Your EMF Health Risks?⏎Good! Learn the one small change you should make right now.⏎To understand this
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
- L8:1848 excites
- L8:200 suppresses
- L6:2621 suppresses
- L7:337 excites
- L8:573 excites
- L8:410 suppresses
- L8:911 suppresses
- L7:2570 excites
- L4:291 excites
- L8:361 excites
- L7:1197 excites
- L8:982 excites
- L4:175 suppresses
- L8:197 excites
- L6:250 excites
- L8:187 excites
- L7:2811 excites
- L3:487 suppresses
- L7:948 suppresses
- L6:2282 excites "and" after a contrast or pause
- L8:444 suppresses
- L8:3048 excites
- L8:441 excites 'or' linking alternatives
- L6:137 suppresses
Feeds later layers
- L13:345 excites
- L17:629 excites question mark after a rhetorical question
- L10:308 excites
- L10:144 excites
- L12:1405 excites
- L10:137 suppresses 'the' before a noun after 'of'
- L19:1072 excites
- L15:1543 excites
- L10:213 excites
- L17:2684 excites
- L13:346 excites
- L17:1567 excites
- L10:911 excites
- L14:1124 excites
- L14:2005 excites
- L17:839 excites
- L10:452 excites
- L17:72 suppresses
- L11:180 excites
- L11:25 suppresses
- L13:918 excites
- L13:1375 excites
- L19:282 suppresses
- L13:9 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
- asn
- (bin
- уй
- şim
- asn
- ному
- setQuery
- ERO
- itim
- 它可以
Suppresses
- into
- inely
- untreated
- 镑
- 除外
- estruct
- 小时前
- 真情
- ellen
- plaintext