'th' in ordinal numbers
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 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 responds to the 'th' ending of ordinal numbers written in digits, such as 84th or 70th. Steering it up or down did not produce text that continued that pattern.
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.
-
125 killed, wounded, or missing. In February 1864, the 84th Indiana camped at Cleveland before joining Gen. Sherman's campaign toward Atlanta in May
-
Army. Indiana Infantry Regiment, 84th (1862-1865)⏎Historical Note: The 84th Indiana Regiment was formed in Richmond, Indiana (Wayne Co.), and mustered
-
of the 328th who felt Sergeant Bernard Early and other members of the patrol deserved credit, too.⏎
-
in Kildare and Offaly. BOYLE is one of 50 common Irish surnames of modern Ireland, as well as the 84th most popular last name in Scotland. Surname Origin: Irish, Scottish Alternate S
-
age. "If we go back to nature, we would survive. Because as age grows, healthy food is important, "said the 1037th UGM doctoral graduate.⏎In conclusion of her dissertation entitled "Biosystematics
-
learning plan template doc, professional learning plan template.<|endoftext|>By: Jay Anderson⏎September 13, 2009 marked the 110th anniversary of the demise of Mr. Henry H. Bliss. Mr. Bliss had
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 question of the mind is a big one. But in the past, it was about the mind, and the mind is about the mind. But this is a big one. The mind is about the mind. But this is a big
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Chuginadak Island of the Aleutian Islands of Alaska. It originates like all its siblings, the volcanic Aleutan Islands due to the Pacific Plate subducting under the North American Plate.⏎The name comes from grover Cleveland, who
-
is described in multiple online sources, as addition to our editors' articles, see section below for printable documents, Ballotechnics books and related discussion.⏎Suggested Pdf Resources
-
procedures; however, all methods of declaw were proven painful. Pain levels were the same six days into recovery regardless of procedure type in one study.5,6 Another study comparing pain from scalpel declaw vs. laser declaw showed very
-
eneration. The age-related disc degeneration in the sand rat has been shown to be reliable and predictable, and has been well characterized in previous radiologic studies [1–9]. The present work extends the previous radiological studies in to develop
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
- L4:2561 suppresses
- L2:1771 weakly suppresses
- L2:2453 weakly suppresses
- L2:167 weakly excites
- L2:1990 weakly excites
- L4:2950 weakly suppresses
- L4:2364 weakly suppresses
- L3:2294 weakly excites
- L2:2786 weakly excites
- L3:1165 weakly excites
- L4:2254 weakly suppresses
- L4:2286 weakly suppresses
- L4:2852 weakly excites
- L2:458 weakly suppresses
- L0:2201 weakly suppresses
- L2:1423 weakly suppresses
- L1:334 weakly suppresses
- L4:1379 weakly excites
- L3:975 weakly excites
- L4:2628 weakly excites
- L3:1900 weakly excites
- L4:2650 weakly excites
- L2:824 weakly suppresses
- L2:2109 weakly suppresses
Feeds later layers
- L25:624 weakly suppresses
- L13:1726 weakly excites
- L21:2852 weakly excites spot where a number or 'time' comes next
- L7:2140 weakly excites
- L12:718 weakly suppresses
- L25:859 weakly suppresses
- L9:1846 weakly suppresses
- L13:412 weakly excites
- L11:1389 weakly suppresses
- L7:1716 weakly suppresses
- L18:514 weakly excites
- L7:1337 weakly suppresses 'of' inside organisation and product names
- L12:1116 weakly suppresses 'of' in formal noun phrases, mixed
- L24:307 weakly excites
- L8:952 weakly suppresses
- L17:2741 weakly excites
- L26:755 weakly excites
- L12:1471 weakly excites
- L18:1070 weakly excites
- L25:1969 weakly excites
- L6:389 weakly suppresses
- L21:3040 weakly excites
- L10:1827 weakly excites
- L24:1362 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
- <meta
- лас
- ASCII
- Porter
- sym
- 每人
- 每一
- omon
- 出租
- sworn
Suppresses
- 료
- xious
- deadline
- ollo
- initView
- 隈
- 佶
- \Blueprint
- 葳
- vỏ