the word before 'risk'
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. 25 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
This neuron fires on the word just before 'risk', as in 'at risk' or 'higher risk'. It appears to be predicting that the word 'risk' comes next. Turning it down made the model repeat 'risk' over and over, which shows how strongly it pushes that word.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
how to care for people with disabilities, it promotes bias in healthcare. As a result, patients with disabilities lose trust in medical care, which can put their lives at risk.⏎The Impact of Low Accreditation Standards⏎On Dec. 14
-
a $195,000 Rapid Response Research grant from the National Science Foundation to, using machine learning, identify which COVID-19 patients are at risk of adverse cardiac events such as heart failure, sustained abnormal
-
, cardiogenic shock and death.⏎Increasing evidence of COVID-19’s negative impacts on the cardiovascular system highlights a great need for identifying COVID-19 patients at risk for heart problems, the researchers say. However, no such predictive capabilities currently
-
in the inner Bulgarian regions. Currently, an active surveillance system is in place for dairy factory employees and persons considered at risk after outbreaks in ruminants.⏎During 2005–2
-
desktop! more...⏎Orthodontics are often necessary to help improve the stability, function and health of an individual's teeth; otherwise, many people would be at higher risk for gum disease, tooth decay and tooth loss because of improper teeth positioning in
-
safely handle their vehicles and make prudent travel decisions. When bus drivers do not adhere to proper bus-handling protocol or traffic laws, they put dozens of vulnerable children at risk. The most common causes of fatal school-bus accidents include the following:⏎
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: risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk risk
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
diligently to perform mitzvot (good deeds) to find favor with God during the 10 days of repentance between Rosh Hashanah and Yom Kippur.⏎According to Jewish teaching, God sits in judgment of the
-
to send the allied forces pell mell towards Berlin and get there before the Russians, Roosevelt was⏎adamant about sticking to the agreed upon plan conceived at the Yalta conference. Eisenhower would have liked ot send his forces to Berlin (Patton
-
J.M.; Mortimer, S.R.; Brereton, T.; Redhead, J.W
-
commemorate the Jewish holiday, Yom Ha’ Shoah, will begin at 5 p.m. in the
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
- L17:2930 excites
- L22:2775 excites
- L19:2152 suppresses
- L23:2614 excites 'at' in 'at risk'
- L21:1439 suppresses
- L22:3060 excites
- L21:1887 suppresses
- L21:820 suppresses
- L18:2623 suppresses
- L15:2782 suppresses
- L21:1045 suppresses
- L20:1678 suppresses
- L21:1472 weakly suppresses
- L20:2973 weakly excites
- L23:1767 weakly excites
- L19:1461 weakly suppresses
- L21:2227 weakly excites
- L21:1699 weakly suppresses
- L19:2157 weakly suppresses
- L14:2691 weakly suppresses
- L22:1767 weakly suppresses
- L19:2900 weakly excites
- L17:3011 weakly suppresses
- L12:1715 weakly excites
Feeds later layers
- L26:1950 suppresses
- L27:221 weakly suppresses
- L26:2346 weakly suppresses
- L27:2273 weakly suppresses
- L27:2882 weakly suppresses
- L27:336 weakly suppresses
- L26:2591 weakly suppresses
- L26:2778 weakly suppresses
- L27:2396 weakly suppresses
- L27:2264 weakly suppresses
- L26:1495 weakly excites
- L26:1326 weakly suppresses
- L26:1314 weakly suppresses
- L27:2538 weakly suppresses
- L27:2695 weakly suppresses
- L26:2286 weakly suppresses
- L26:2205 weakly excites
- L26:2630 weakly suppresses
- L27:1759 weakly suppresses
- L26:2455 weakly suppresses
- L27:1630 weakly suppresses
- L26:1276 weakly suppresses
- L27:1191 weakly suppresses
- L26:627 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
- ")).
- "));
- "));⏎
- ")); ⏎
- "));⏎⏎
- ")));⏎
- ")));
- defgroup
- ades
- ."));⏎
Suppresses
- risk
- 风险
- risk
- Risk
- Risk
- 的风险
- 風險
- risks
- -risk
- 风险管理