'to' after 'have got' or 'will have' (obligation)
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. 4 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 'to' when it follows 'have', 'has got', or 'will have' and introduces something that must be done, like 'will have to move' or 'has got to be created'. It is a clear and consistent pattern. When turned up, the model sometimes wrote about needing to do things, though this was not consistent.
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.
-
to be at a disadvantage for heat loss,” Hindle explained. “They’re more likely to lose heat to the environment, and therefore they’re going to have to ramp up heat production.”⏎The three-year grant from the National Science Foundation’s Division
-
. Even if you don't make it all the way to the top of the mountain or hike all 40 miles of the trail, you'll have to hike just to get back to civilization. Unless you quit within the first mile or
-
solution to increase food and could be one of the next most important developments in human history. By 2050, the edible bioresource biomass will have to satisfy the 9 billion people predicted to live on the planet. Seaweed
-
that you can have whatever you want whenever you want without having to work for it. It is about being convinced that you deserve things you want and someone has got to give them to you no matter what.⏎Below are what I believe are some of
-
A checklist for evaluating would-be donors and fully explaining the risks — plus requirements to better monitor those donors' health and social stability for two years. Centers would have to track such
-
, adaptable, and easily taught.⏎“The social relationships [between] these things matter a lot. And these things have got to be created, developed, made to be sustainable as much as the tech,” R
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 number of a certain type of object, and the number of other types of objects in the world, and the world is known as the number of the people. The world is the number of the people, and the number of the people
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
report:<|endoftext|>Marketing is related to the exchange of goods and services. Through its medium the goods and services are brought to the place of consumption. This satisfies the needs of the customers. The following activities are undertaken in respect of the exchange of goods
-
buddy’ and ‘bro’ in the OED have recently been revised. We explore their history and increase in popularity.<|endoftext|>NEED FOR ASSISTANCE⏎People with autism needed assistance with a range of activities, with
-
help the students to develop and understand the concept of child health, the trends in child care and health problems of children. It will enable the students to meet the needs of the children, in health and sickness.⏎This course book is designed for students
-
develop a database for historical analysis. It provides an ingenious and elaborate instrument to process, record, and analyze any type of historical data and sources. Its use demands a
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
- L0:2662 excites
- L3:266 weakly suppresses
- L3:1787 weakly suppresses
- L2:2739 weakly suppresses
- L2:449 weakly excites
- L3:2760 weakly excites
- L2:2918 weakly suppresses
- L0:1592 weakly suppresses 'Springer' and 'rebirth' style word pieces
- L2:1494 weakly suppresses
- L2:2640 weakly suppresses
- L0:1671 weakly suppresses
- L1:1641 weakly suppresses
- L3:2822 weakly excites
- L1:2801 weakly excites
- L2:1548 weakly suppresses
- L2:126 weakly excites
- L1:2173 weakly suppresses
- L1:2497 weakly excites
- L2:2328 weakly suppresses
- L1:2127 weakly suppresses
- L3:2843 weakly excites
- L3:50 weakly excites
- L0:2834 weakly excites
- L3:2154 weakly suppresses
Feeds later layers
- L16:2349 strongly excites
- L10:520 excites
- L17:2345 excites
- L9:640 excites
- L17:1822 excites 'should' in advice and claims
- L16:2563 excites
- L5:2538 excites
- L17:451 excites
- L6:2182 excites
- L10:1576 excites
- L7:1232 excites
- L14:2124 excites
- L21:2252 excites
- L10:562 excites
- L7:3004 excites
- L5:1846 excites
- L21:3062 excites
- L5:1701 excites
- L20:2529 excites
- L24:2712 weakly excites
- L18:2568 weakly excites
- L8:1465 weakly excites
- L20:2395 weakly excites
- L25:1092 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
- обязательно
- 不得不
- obliged
- must
- слав
- require
- ACM
- 必须
- opers
- 就必须
Suppresses
- représent
- etable
- สบาย
- uation
- setMessage
- tings
- numeros
- ispers
- setResult
- uations