'acid' and related word parts
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. 20 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 'acid', especially in words like 'acidophilum', 'acidifies', 'acid rain' and 'acid based'. Turned up, the model produced number lists and repeated years, so the steering output does not show the same 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.
-
11 November by the Molecular Systems group at the LMS shared insights into these questions.⏎Building chromatin-like structures⏎The archaeon Thermoplasma acidophilum grows optimally in very acidic conditions with high temperatures. It doesn’t
-
the usage of plastic bags, plastic straws and switch to a compostable products and reuse those items that are reusable.<|endoftext|>Sixty-three years ago the first acid trip was taken by an unwitting research chemist, Albert Hofmann, who has
-
hydrologic consideration.⏎—Pervious concrete, water purification, acid rain, waste motor oil.⏎Ming-Ju Lee is with the Jen-T
-
we’ve harvested.⏎Now it’s time to go plant a reef.⏎I got here with a question: What happens to coral reefs as the ocean warms and acidifies over the coming century? Corals might seem removed from human society, but
-
's has introduced several preprocessing methods to reduce the log data size and to eliminate incorrect records, which lead to improved quality, quantity and performance.<|endoftext|>Based on the acid rain and concurrent meteorological observational data during the past 10 years in Fuj
-
Qualitative Crackle test is used for engine oils.)⏎- Acid Number measures acid buildup, which denotes oxidation.⏎- FTIR measures chemistry changes in a lubricant
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 first check to check the first check, the first check to check the first check... The first check to check the first check, the first check to check the first check... I am aware that the first check check check check check check
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
secure digital communications. Specifically, if large quantum computers can be built then they may be able to break the asymmetric cryptographic primitives that are currently used in TLS, the security protocol behind HTTPS.⏎Quantum computers exist today but, for the moment, they
-
, without safety screening⏎Reproduced with permission from Fluechtling.net⏎- Millions
-
the adhesive or sealant to take advantage of the microscopic roughness etched to the surface by atmospheric plasma, locking them together. In<|endoftext|>Here at CRC, safety is our number one priority. We have dedicated staff who can offer a range of
-
advances in a decade.⏎Rechargeable lithium-ion batteries became costly for General Motors Co. (GM) after they caught fire in its plug-in Volt car during safety tests in 2011. They grounded Boeing Co. (BA)
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
- L19:407 excites
- L4:1761 suppresses
- L19:314 excites
- L18:287 excites
- L4:2347 weakly suppresses
- L2:852 weakly suppresses
- L4:741 weakly excites
- L4:834 weakly suppresses
- L5:800 weakly suppresses
- L0:2907 weakly suppresses
- L4:428 weakly suppresses
- L0:2468 weakly suppresses
- L19:256 weakly suppresses
- L17:650 weakly excites
- L18:519 weakly excites
- L5:1946 weakly suppresses
- L18:1069 weakly excites
- L8:500 weakly excites 'to' after verbs like continue or help
- L6:105 weakly excites
- L13:743 weakly suppresses
- L10:299 weakly suppresses
- L0:3031 weakly suppresses
- L18:2360 weakly suppresses
- L15:2804 weakly excites
Feeds later layers
- L23:670 suppresses
- L25:1401 weakly excites
- L25:1890 weakly suppresses
- L24:2737 weakly excites
- L22:2584 weakly suppresses
- L24:886 weakly excites
- L27:2375 weakly suppresses
- L24:1112 weakly suppresses
- L22:775 weakly suppresses
- L24:1775 weakly suppresses
- L21:1116 weakly suppresses
- L23:808 weakly suppresses
- L24:2500 weakly suppresses
- L21:1887 weakly suppresses
- L22:229 weakly suppresses "of" after a noun in "X of Y"
- L22:2477 weakly excites
- L21:2535 weakly excites
- L21:2814 weakly suppresses
- L24:841 weakly suppresses
- L23:996 weakly suppresses
- L23:1227 weakly suppresses
- L23:2081 weakly suppresses
- L27:2725 weakly excites
- L23:1592 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
- inea
- аль
- igm
- ighet
- ifiers
- gency
- alendar
- ifferential
- als
- äl
Suppresses
- 系数
- �数
- preocup
- HECK
- scares
- 饵
- checks
- ECH
- vos
- poids