the 'therm' stem in thermo- words
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. 23 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 start of words built from 'therm', such as 'thermometer', 'thermionic' or 'thermophilic', which are about heat. Turning it up gave replies about temperature and heat in a few cases, but many were unrelated.
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.
-
out more.⏎1. Maintain the Fridge Temperature⏎A recommended fridge temperature of 5 degrees C is advised. There are thermometers available at reasonable price which can help you determine the temperature. A report suggests
-
Roman god who served as messenger for the other gods. The silver-white chemical is the only common metal that is a liquid at room temperature. It is used in thermometers,
-
today it still features in commercial and domestic microwave ovens seven decades after its introduction. In fact, the magnetron belongs to a broader Class of products known as thermionic valves, or ‘tubes’. Valves were relegated to specialist applications once
-
; the indirect techniques only calculate it from other indirect indicators.⏎Image Credit: Soonthorn Wongsaita / Shutterstock⏎Many devices for LOD measurement use the thermogravimetric principle. This uses the total loss of weight incurred by the
-
the machine configuration file for assembly binding, see How the Runtime Locates Assemblies.⏎An application configuration file contains settings that are specific<|endoftext|>Mercury is used in thermometers, fluorescent light bulbs, and some electrical switches.⏎"Mercury" is
-
use of epoxy-based resins and may be classified as the thermosetting epoxide polymer. Exteriors that receive this coating are resistant to
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: at some moment, at least one of the two atypoeyypierym entries at the more than two than two compared to atypoeyy atyypierym atyypierymat compared to atypo
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, Mithra, was adopted and developed by the Zoroastrians as an angelic
-
no ultimate evil, only delusions represented by demons, who need to be guided back on the Buddhist path. Represent<|endoftext|>Tam Nông is a district of Đồng Tháp province, part of the once very vast area called Đồng Tháp Mười
-
rule in Britain at the beginning of the fifth century.⏎|DEO L SENTIVS CASTVS > LEG VI D P|⏎|"For the god [Mithras], Sentius Castus, centurion of the Sixth Legion makes this
-
Guisers (children disguising themselves at Halloween) in Scotland sang:⏎If ye dinnae let us in,⏎We will bash yer windies in.⏎"Whuppity Scooorie in Lanark is a festival, believed to have survived
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
- L4:52 excites
- L16:792 excites "the" inside a prepositional phrase
- L22:1882 excites
- L17:30 suppresses
- L20:361 weakly excites
- L18:799 weakly excites
- L0:87 weakly excites
- L3:2846 weakly excites
- L19:1064 weakly suppresses
- L20:792 weakly excites
- L16:419 weakly excites
- L10:1210 weakly suppresses
- L18:648 weakly suppresses
- L19:472 weakly excites
- L15:2129 weakly suppresses
- L9:24 weakly excites "the" before a named organisation or title
- L21:614 weakly excites
- L20:211 weakly excites
- L7:122 weakly excites
- L22:1044 weakly excites
- L5:1952 weakly excites
- L2:2780 weakly excites
- L20:1079 weakly suppresses
- L20:894 weakly excites
Feeds later layers
- L25:3011 weakly suppresses
- L27:1215 weakly suppresses
- L27:1100 weakly suppresses
- L24:223 weakly suppresses
- L27:599 weakly suppresses
- L24:1627 weakly suppresses
- L24:13 weakly excites
- L26:2914 weakly excites
- L26:577 weakly suppresses
- L26:1637 weakly excites
- L25:2325 weakly suppresses
- L27:359 weakly excites
- L24:911 weakly suppresses
- L25:501 weakly excites
- L24:2722 weakly suppresses
- L25:656 weakly excites
- L26:2182 weakly suppresses
- L25:2774 weakly excites
- L25:1172 weakly excites
- L27:731 weakly suppresses
- L26:1915 weakly excites
- L26:2621 weakly suppresses
- L24:2187 weakly excites
- L25:850 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
- Liberation
- tle
- ).__
- rello
- ervention
- =count
- Means
- 就是在
- NEL
- ADI
Suppresses
- ew
- yst
- irst
- er
- ered
- wards
- erb
- 頭
- iei
- ir