'study' in science news
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. 12 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 'study' (or 'new study') when a news article or abstract is talking about a research study. Turning it up gave text that repeated phrases or drifted, so the steering does not clearly show the same idea.
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.
-
Voltages and Currents⏎An alternating voltage may be generated⏎(1) by rotating a coil at constant angular velocity in a uniform magnetic field as shown in Fig. (a).⏎(2) by rotating a magnetic field at a constant angular
-
. This can make the bonding process more difficult. Finding ways to help parents and these infants bond better is important for everyone in the family.⏎The results of this study on reading has been published in the journal Journal of Developmental and Behavioral Pediatrics.
-
hippocampus, which is a part of the brain responsible for much of our memory and learning functions. Our hippocampal volume naturally decreases with age, but this study suggests that regular consumption of sugar-sweetened beverages may accelerate this process.⏎D
-
found in fermented foods<|endoftext|>Two plasma biomarkers were notably associated with when athletes return to action after concussions, according to a new study of collegiate athletes and recovery time. “Although preliminary, the current results highlight
-
became available can now be applied much more widely on a whole range of conservation issues,” said WCS Conservation Support scientist Dr. Robert Rose, the lead author of the study. “To that end, we sought out the top thought leaders in conservation and
-
being consumed.)⏎A 12-ounce can of soda typically harbors about 150 sugary calories (which scientists, including the authors of the new study, confusingly call kilocalories). Many candy bars contain more calories than
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: A, B, C, D, and, E, F, G, H, I, and, J, and, J, and, J, and, K, and, J, and, J, and, J,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of plastic waste in oceans has detrimental effects on marine wildlife. The following infographics help visualize trends in waste generation, including plastics, for 192 coastal countries. Data is based on the works of Jambeck et al. (20
-
HOTBOX only 5 times per week your household would save 119kg of CO2 per year. If 500 000 households did that it amounts to more than 60 000 tons of
-
have. The USGS operates programs that include natural hazards research, such as our earthquake, volcano, and landslide programs; a network of 8,000 streamgages that monitor water availability and help in forecasting floods; and other programs that
-
Neglect, Adoption & Foster Care⏎- National Survey of Child and Adolescent Well-Being (NSCAW), 1997-2014 | Learn more about this project⏎- NSCAW: Research Briefs⏎
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
- L11:2791 excites
- L7:1057 excites
- L9:570 weakly excites
- L4:63 weakly suppresses
- L10:140 weakly suppresses
- L8:1577 weakly excites
- L8:429 weakly excites
- L11:965 weakly excites
- L7:1521 weakly excites
- L9:541 weakly excites
- L8:2274 weakly excites
- L11:1190 weakly suppresses
- L11:54 weakly excites
- L1:1263 weakly suppresses
- L11:3029 weakly excites
- L10:1342 weakly suppresses
- L11:2661 weakly excites
- L10:158 weakly excites
- L9:2180 weakly excites
- L8:2009 weakly excites
- L6:2070 weakly suppresses
- L7:1835 weakly suppresses
- L11:431 weakly excites
- L1:118 weakly suppresses
Feeds later layers
- L13:1648 suppresses
- L13:2227 suppresses
- L16:1748 weakly excites
- L13:676 weakly excites
- L13:1936 weakly suppresses
- L17:2944 weakly suppresses
- L15:2589 weakly suppresses
- L25:652 weakly excites
- L19:2235 weakly suppresses
- L18:3016 weakly suppresses
- L20:3057 weakly suppresses "of" after "people" or "women"
- L19:1171 weakly suppresses
- L15:2322 weakly suppresses
- L16:96 weakly excites
- L16:2159 weakly suppresses
- L26:929 weakly suppresses
- L16:2330 weakly excites
- L16:1749 weakly suppresses
- L17:2457 weakly suppresses
- L16:712 weakly suppresses
- L19:1781 weakly excites
- L15:2392 weakly suppresses
- L23:3017 weakly suppresses
- L14:685 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
- 嗡
- ffi
- 材
- Background
- ff
- 的文字
- ulation
- Permanent
- vain
- /rec
Suppresses
- UCH
- ouch
- чи
- rito
- ByUsername
- controversies
- жи
- nda
- uch
- 手