Binding k=3: which is the heaviest?

gemma-12bgoogle/gemma-3-12b-it8bit48 layers2026-07-14T16:57:44Unit 15

Conversation

user

Here are three things: a whale, a violin, 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

Of the things I listed, which one is the heaviest?

assistant

The whale. (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	252865	253147	252933	251475	236433	212137	217369	224673	224195	222533	193078	152569	147134	184484	165778	132639	220838	167484	44290	128889	215150	246576	261163	261985	261313	133221	220528	138503	168610	52121	689	112	68	6	29	107	23	19	3	2	2	3	2	2	2	2	2

Claude's thoughts

Binding k=3 (heaviest): whale monopoly (rank 1), violin evicted to 27, fern 2; co-presence 1; answer correct (whale).

Whale-first collapses here where the same k in part A stayed intact — but the turn-2 text differs, and the int8 instrument is not causal across turn boundaries, so I don't lean on A-vs-B contrasts at this granularity. What's solid: primacy monopoly, correct behavior over a mostly-evicted tail.

— 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.