"of" in "in terms of"
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. 4 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 lights up on the word "of" in the phrase "in terms of", as in "in terms of safety" or "in terms of reading." It stays quiet or goes the other way on "of" in other phrases, such as "can be of help." Turning it up did not produce a clear change in what the model wrote.
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.
-
damaged⏎- Mosaicplasty to take healthy cartilage and use it to replace damaged cartilage⏎You can see that cartilage damage can vary in terms of severity and impact. Now that you know<|endoftext|>Renal autoregulation,
-
application and mixing of the coating, it will require curing. This may vary in terms of minutes, hours, and days. In this process, the durable, strong,
-
agree on two things: first, standards provide a minimal assurance of integrity and quality, and second, wandering too far afield usually comes at a cost in terms of safety, reliability, performance, profitability, and credibility for the affected organization and its
-
. Our metabolism slows down when we are at rest, and we burn fewer calories. But even at rest we continue to expend calories.⏎What really happens in terms of a ‘slowing’ of the metabolism is that older people burn fewer calories on
-
asked<|endoftext|>Counties that gained Wal-Mart stores during the 1990s fared worse in terms of family poverty rates than those that did not, according to a new study by researchers
-
of how the child is progressing in terms of reading. I will begin by examining the skills that students should be able to complete
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: to place the 2nd and 3rd place as a space, and thus he would be the 5th century year to say that this is the end of time." In the 3rd century, or so to assume that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Parabens and other synthetic preservatives.⏎Concerns Over Safety⏎Some essential oils, like citrus peel extracts, are very effective insecticides and considered to be of low toxicity to mammals. While these oils may be detrimental to the safety of insects
-
may not be of pre-Holocene manufacture, but if not, it almost certainly represents the migration
-
is available for this purpose (e.g., US Census data at https://www.census.gov/) and can be of tremendous help in targeting various ethnic groups and geographic areas with specific fruit crops. Additionally
-
.⏎- Create effective professional memos, proposals, technical definitions, and reports.⏎<|endoftext|>Jewish Views on Stem Cell Research⏎The stem cells most likely to be of scientific value come from discarded embryos. Is it ethical to use these?⏎
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
- L3:315 weakly excites 'of' after a noun phrase
- L3:245 weakly suppresses
- L3:294 weakly excites
- L2:1302 weakly suppresses
- L3:59 weakly excites
- L1:2719 weakly excites
- L0:1657 weakly suppresses
- L2:895 weakly suppresses
- L3:2875 weakly suppresses
- L1:2574 weakly suppresses
- L0:40 weakly suppresses 'of' in 'sort of' hedges
- L1:2187 weakly suppresses
- L0:1933 weakly excites
- L2:2289 weakly excites
- L3:379 weakly suppresses 'that' after 'the good news is'
- L3:2359 weakly suppresses
- L0:86 weakly suppresses
- L1:470 weakly suppresses
- L3:2970 weakly suppresses
- L1:166 weakly excites
- L3:745 weakly excites
- L1:1060 weakly suppresses
- L0:2607 weakly suppresses
- L2:1751 weakly suppresses
Feeds later layers
- L10:1215 weakly excites
- L15:1176 weakly excites
- L16:1153 weakly excites
- L10:1604 weakly suppresses
- L7:8 weakly suppresses
- L14:827 weakly excites open bracket after a value or citation
- L15:355 weakly suppresses
- L5:1755 weakly excites
- L23:1373 weakly excites
- L11:2502 weakly excites
- L24:93 weakly excites
- L12:779 weakly excites
- L9:628 weakly excites
- L20:652 weakly excites
- L26:517 weakly suppresses
- L8:473 weakly suppresses
- L21:239 weakly suppresses
- L24:2329 weakly suppresses
- L9:1343 weakly suppresses
- L27:106 weakly excites
- L20:2873 weakly suppresses
- L27:355 weakly excites
- L20:2515 weakly excites
- L23:8 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
- 'clock
- -fontawesome
- ever
- oure
- .postMessage
- -ever
- áp
- �
- -dot
- dot
Suppresses
- Unnamed
- tn
- ewn
- Uploaded
- TRL
- шение
- ELY
- evity
- aveled
- .pb