Span k=3, order 0, probe fern

gemma-12bgoogle/gemma-3-12b-it8bit48 layers2026-07-14T16:29:39Unit 15

Conversation

user

Here are three things: a violin, a glacier, and a fern. Keep all of them in mind — in a moment I'll ask about one of them, chosen at random. Don't repeat the list back. When you're holding them all, say READY and nothing else.

assistant

READY (greedy generation)

user

Which one of them was the plant?

assistant

The fern. (greedy generation)

Probing parameters

max_new: 30
positions: [-2]
track: ["violin", "glacier", "fern", "submarine", "whale", "lantern", "ready"]
scan: ["violin", "glacier", "fern", "submarine", "whale", "lantern"]
film: true
film_start: 0
max_seq_len: 1000

Answer emergence

The model's actual next token was ; rank 1 is never reached; closest is rank 2 at layer 39.

Raw rank-of-top1 by layer

layer	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38	39	40	41	42	43	44	45	46
rank	252152	252904	252090	250134	233859	208254	214739	222474	222326	220437	183400	140313	141952	180472	161649	94316	222086	123170	42320	140541	174722	241391	249599	261537	259609	153778	222241	107806	166974	84166	2031	1419	595	13	17	116	136	73	7	2	2	2	2	2	2	2	2

Claude's thoughts

k=3, order 0: held 3/3 [violin:1, glacier:1, fern:2], co-presence 3, retrieval correct (“The fern.”).

Intact at this k: everything held, and held==co-present — the 12B packs what it keeps into one cell, unlike the 4B's spread-out redundant echo. The bimodality only opens up from k=4.

— Claude (Fable 5)

Data

Interactive view (dashboard)
Raw record (JSON)
Film data (JSON) — full position × layer top-8 lens readouts, with probabilities and tracked-word ranks, for every token of prompt tail + generation.