'instead of' and 'rather than'
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. 17 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 lights up on the word that follows 'instead' or 'rather', such as 'of', 'than' or 'try', when a sentence sets one way of doing something against another. Turning it up did not make the output show this idea, though, so the steered text does not clearly reflect it.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
<|endoftext|>The move to electronic health records is in full swing for most of the medical community, but holdouts point to security issues as an argument to keep their records on paper.⏎The news is riddled with stories about data being compromised these days,
-
the stage where you use technology to substitute what you could have done with pen and paper. one popular example of this is using word processor to type a story instead of handwriting it.⏎This is where technology is used to carry out learning tasks in relatively
-
is a distributor as well. Jeffrey Humphries is the contact that I know of for us Canadians, eh.⏎Breakout EDU has gone Digital too! Rather than having kids work on their own
-
example, your tomato ketchup or mayonnaise⏎- Stop using plastic drinking straws and instead try metal, paper, or even wheat straws⏎- Swap tampons and pads
-
your face. Do not forget to clean your headphones regularly.⏎- Use voice commands⏎- Nowadays, AI is becoming more intelligent. We can use voice commands instead of typing on the screen.⏎- Avoid sharing your devices with others⏎- If unnecessary
-
is transported in containers. Worldwide, more than 400 ports have the capability to handle containers.⏎Many ocean carriers are now shipping cargoes in containers rather than break bulk. Generally<|endoftext|>CDPATH is an environment variable
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 purpose of this is only a single purpose, and what is the purpose of this is only a single purpose is only a single purpose, and the purpose of this is only a single purpose is only a single purpose is only a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
be ruled out. A step-by-step approach should be used to identify a possible DISR:⏎1. Complete a drug history. Reconciliation of all prescription and nonprescription agents, as well as any natural medicines, should be completed. Develop
-
to have actinic keratoses. The incidence is slightly higher in men, because they tend to spend more time in the sun and use less sun protection than women do. African-Americans, Hispanics, Asians and others with darker
-
Engineering. “This is one area of stem cell biology where human and mouse differ significantly and we never would have discovered this if we had limited our studies to using only mouse embryonic stem cells. Adult human stem cells just didn’t work for this
-
here.⏎If you would like to support the mission of Open Culture, consider making a donation to our site. It’s hard to rely 100% on ads, and your contributions will help us continue providing the best free cultural and educational
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
- L15:2838 suppresses
- L16:2561 suppresses
- L16:697 suppresses
- L12:1954 suppresses 'with' or 'from' after comparison words
- L16:2278 suppresses
- L12:1834 suppresses
- L11:1215 suppresses
- L12:2471 suppresses
- L16:793 excites
- L13:634 excites
- L14:1413 excites
- L14:172 excites
- L14:1178 suppresses
- L16:2252 suppresses
- L13:2269 suppresses
- L10:2736 weakly excites
- L15:2057 weakly excites
- L1:1082 weakly suppresses
- L9:952 weakly excites
- L6:1386 weakly excites
- L16:1616 weakly excites
- L16:2986 weakly suppresses
- L5:2269 weakly excites
- L10:2924 weakly suppresses
Feeds later layers
- L18:2433 excites
- L19:1893 excites
- L22:1497 excites
- L26:2116 excites
- L21:886 weakly suppresses
- L23:871 weakly excites
- L21:3000 weakly excites
- L25:943 weakly excites
- L24:2188 weakly excites
- L19:2897 weakly suppresses
- L19:1408 weakly excites
- L18:1423 weakly excites
- L20:618 weakly suppresses
- L22:1207 weakly excites
- L19:2484 weakly suppresses
- L21:1932 weakly excites
- L22:1720 weakly excites
- L23:2358 weakly excites
- L18:1892 weakly suppresses
- L25:2846 weakly suppresses
- L18:255 weakly suppresses
- L20:1132 weakly suppresses
- L20:1525 weakly excites
- L20:1920 weakly excites
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
- 手工
- 手动
- manually
- raw
- (raw
- 直接
- directly
- manual
- 亲手
- direct
Suppresses
- 水量
- rends
- 溶解
- しよう
- intentions
- 誉
- ?\
- imbus
- 治理体系
- ーズ