unclear: planned or deliberate wording
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. 5 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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
The top examples are mixed, with a loose thread around words like 'planned', 'deliberate', 'intention' and 'voluntary', plus land plots and lots. It is hard to say what the neuron is really about. Turning it up or down gave text with no clear link to these words.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
productive corn field.⏎Two town site promoters, Ace Borger and J.T. Peyton, were engaged. The land was then surveyed, subdivided into blocks, lots, streets, avenues, and alleys. The Stinnett Town site Company
-
murdered there, including 960,000 Jews.⏎Auschwitz-Birkenau has become the symbol of the Holocaust and of willful radical evil in our time.⏎What the liberators found in these camps astounded
-
of areas as part of our planned SOW.⏎Some key areas are outlined below:⏎At KS3 we cover modules
-
the Second World War. By adopting this permissive approach to the right to asylum, the Parliamentary Council (Parlamentarischer Rat) wanted to make a deliberate break with the National Socialist past, which had produced millions of deaths, refugees and
-
depicted onto the lot identification plan. Recently, the Land Registry proposes the Land Titles Bill by the replacement of current deeds registration system. However, such proposal has no intention to deal with the long neglected land rights: boundary extent.”⏎Hong Kong represents a
-
plan has been in review status for almost 2 years and it appears that DOE's Office of Nuclear Energy has no intention of finalizing the plan.⏎Recommendation: To ensure that an RTG infrastructure
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 do with this, I am. I am. I need to know. I am. I need to know. I need to know. I need to know. I need to know. I need to know
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
as many base search paths as you like. Here's one that includes⏎/etc as a search path too.<|endoftext|>Salivary gland (Endocrine system)⏎: #1 is , #2 is , #3 is⏎The salivary
-
March 10, 2004⏎: The Chilkoot Brown Bear Project⏎I uttered the habitual salutation as I
-
second voyage to the New World the French arrived off the coast of Florida on Thursday, June 22, 1564 around three or four o’clock in the afternoon. They landed about thirty leagues south of the St. Johns
-
teacher who just lately switched into teaching grade six, I have been on the hunt for middle college teacher sources for the final handful of months. PhD in Unique Education : PhD in Unique Education delivers an write-up about how to assistance students with specific desires
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
- L4:896 suppresses
- L2:280 excites
- L3:75 excites
- L0:2137 weakly excites
- L1:111 weakly suppresses
- L1:1370 weakly suppresses
- L2:742 weakly suppresses
- L0:574 weakly suppresses
- L4:676 weakly suppresses
- L0:643 weakly excites
- L4:2470 weakly suppresses
- L3:2952 weakly excites
- L3:2923 weakly suppresses
- L3:1903 weakly suppresses
- L3:2114 weakly suppresses
- L4:2139 weakly suppresses
- L2:1910 weakly suppresses
- L2:10 weakly suppresses
- L3:2656 weakly suppresses
- L3:2198 weakly suppresses
- L3:248 weakly excites
- L4:1881 weakly excites
- L4:235 weakly suppresses
- L4:330 weakly excites
Feeds later layers
- L9:909 suppresses
- L20:2537 weakly excites
- L6:616 weakly suppresses
- L13:210 weakly suppresses
- L11:300 weakly suppresses
- L22:1163 weakly excites
- L16:1662 weakly suppresses
- L16:1447 weakly suppresses
- L7:26 weakly excites
- L27:1498 weakly excites
- L22:1118 weakly excites
- L10:217 weakly excites
- L23:254 weakly suppresses
- L10:1866 weakly excites
- L17:989 weakly suppresses
- L21:2112 weakly excites
- L21:2451 weakly excites
- L11:2549 weakly suppresses
- L19:1063 weakly excites
- L10:457 weakly excites
- L20:3049 weakly excites
- L25:38 weakly suppresses
- L22:1352 weakly excites
- L19:241 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
- 总局
- descended
- 簠
- atat
- riff
- ;}
- GOR
- ellers
- JsonSerializer
- ⟫
Suppresses
- ront
- .splitext
- 凉
- ternal
- 力还是
- ponents
- �
- ัน
- ceptors
- �