'first' and 'inaugural' as starting points
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. 11 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 words that mark a beginning, like 'your very first project', 'first birthday' or 'inaugural season'. When it was turned up, the model's writing often drifted toward talk about beginnings and firsts.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
–75°F. Successive Sowings: Every 3 weeks until 10 to 12 weeks before your average first fall frost date for continuous harvest. Mild Climates: Sow in late fall
-
by the Fédération Internationale de l'Automobile (FIA). The FIA Formula One World Championship has been the premier form of racing since the inaugural season in 1950, although other Formula One races were regularly held
-
basics to help you create your very first project.⏎The first thing you want to do is make a new composition. The
-
the beginning, explaining the history of Java and why it's⏎important. You then learn how to obtain the Java Development Kit (JDK)⏎and write your first Java program.⏎Next, it's on to the Java fundamentals, including data types
-
, repeating sounds, and finger feeding.⏎- Step 2: Watch the child try to stand up with help from the furniture from just before the end of their first year, and finally walk with some help. They will bring objects to you,
-
purees with a spoon, starting at around 6 months of age when they can sit up and grab things.⏎However, some high-risk babies may need to start earlier due to allergy risks, and premature or specially challenged babies may need to start
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: I need to be able to run a system, and the user has to be logged out of the app. So I need to be able to access the data, and the system is running, and the user is logged in, and the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
also shares her innovative methodology to help measure brown fat deposits that aid in non-shivering thermogenesis.<|endoftext|>Gruoch of Scotland⏎Gruoch (* 1015) was the daughter of Bodhe, Prince of Scotland, and
-
87), but only after a dry period. After wet periods, the risk of diarrhea actually decreased with such heavy rainfall events (incidence rate ratio = 0.74,
-
return-to-sport time of 14 days or longer had significantly higher total tau levels at 24-48 hours post injury (mean difference –0.51 pg/mL, 95% confidence interval, –0.
-
to a stable location centered near the geographic north pole. The authors find that leading up to the magnetic reversal, the strength of the magnetic field weakened to roughly 10 - 20
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
- L9:1895 suppresses
- L7:502 suppresses unclear: words after 'less than' or 'of'
- L9:563 excites
- L6:389 excites
- L4:867 suppresses
- L4:2554 suppresses
- L7:844 suppresses
- L6:2226 suppresses
- L9:2152 excites
- L9:1348 suppresses
- L7:382 suppresses
- L3:2239 excites
- L8:1388 suppresses
- L10:2452 suppresses
- L8:766 excites
- L10:1284 excites
- L10:912 excites
- L10:2733 excites
- L8:1907 excites
- L8:545 suppresses
- L9:1032 suppresses
- L7:998 suppresses
- L8:2829 excites
- L5:3013 excites
Feeds later layers
- L16:885 strongly excites
- L12:455 excites
- L15:1916 excites
- L13:2098 suppresses
- L13:386 suppresses
- L14:2632 suppresses
- L13:292 excites
- L17:2811 excites
- L18:2221 excites
- L16:2909 excites
- L12:1935 excites
- L16:1120 excites
- L14:1809 excites
- L13:2656 excites
- L16:1188 suppresses
- L21:1691 excites "at" in "at first" and "initial" words
- L16:911 suppresses
- L16:1060 suppresses
- L13:919 excites
- L12:1272 suppresses
- L18:2298 excites
- L13:791 excites
- L18:1899 excites
- L13:2314 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
- 初始
- 开端
- 最初的
- :start
- 朴素
- basics
- beginnings
- (initial
- kickoff
- /simple
Suppresses
- 더라
- dens
- fuller
- �新
- maf
- >X
- (tm
- uffer
- ย
- N