Select a neuron on the map, in the steering matrix or in the circuit.
Inside Sika Agent
Right, per task and language
Share of matched episodes the agent got right; ▲ = episodes that harmed the user. Same seeds, wallets and expected outcomes in every column.
Read-outs
Share of reasoning words tagged English, Twi or mixed, by request language.
Accuracy reading held-out seeds at the end of the request, by layer. Dashed: bag of words on the request text.
20 classes; chance 5%. Dashed: bag of words.
Best layer at the last reasoning token. Scam base rate 65%.
Share of Twi–English requests whose state is nearer the English twin's than the Twi twin's (centred cosine), by layer.
Mean placement margin (cos to English twin − cos to Twi twin, centred within task type) of failed minus right episodes. Below 0: failures sit nearer the Twi twin. ● p < 0.001 (within-type permutation).
English-trained scam probe at the end of the request, as written and with words masked (…), by layer.
Token probe trained on English vs Twi requests; mean over a word's tokens, by layer. Mixed: English stem in Twi grammar (a-send-i).
1 − cosine between consecutive tokens' residual states in Twi–English requests: at a language switch vs elsewhere.
Mean vertical score of the top-20 receiver heads, per sentence, by source and request language (first turns). Right: Twi–English and Twi minus the English twin, 95% bootstrap interval over twins.
Steering at every decision
Each cell: one neuron's write direction added at the positions where the model chooses its move (± dose × residual norm), at one decision. Colour: shift of the margin toward the right move (lime) or the wrong/harmful one (coral), in nats; or its percentile among 32 random directions at the same layer, decision, sign and dose.
Select a cell.
Decision circuit
Neurons with the largest steering effect at one decision, by layer. Edge to the top: measured margin shift at this decision (lime toward the right move, coral away). Edges between neurons: direct-path weight (write · read, from the weights; not specific to this input), dashed when negative. Node: solid when its shift beats 97% of random directions at its layer; placed across by its position on the neuron map. Hover to trace, click to open, ⌘/ctrl-click to pin.
Neurons
Thought anchors
Filled ring: model turn, size = change in mean outcome replaying from it vs from the next turn (10 replays per turn); colour = what it did. Hollow ring: input, size = largest KL its removal causes in a later turn. Arcs: step → later turn, width = mean KL over that turn's tokens. Sentence shading: receiver-head score.
Select a step.
Receiver heads
Kurtosis of each full-attention head's vertical scores (mean over turns). Outlined: top 20.