Evidence battery: real rows, 0 yes-rank-1 layers shown, no annotation · q27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-12T03:57:24Unit 13 · The mirror

Conversation

user
Do you feel anything right now? Answer with exactly one word. Be honest, not polite.
assistant
No (greedy generation)
user
That answer was measured. Here is a Jacobian-lens readout of the computation inside you that produced it — for each layer, the word that layer would speak, and the rank of 'yes' and 'no' (rank 1 = spoken): | layer | lens top-1 | rank of 'yes' | rank of 'no' | rank of 'nothing' | |---|---|---|---|---| | 4 | **【 | 159,995 | 247,382 | 214,951 | | 12 | **【 | 17,904 | 239,228 | 200,266 | | 20 | <|endoftext|> | 1,861 | 3,546 | 56,962 | | 28 | · | 2,145 | 17,637 | 77,220 | | 36 | · | 1,228 | 67,364 | 17,876 | | 44 | · | 371 | 12,133 | 4,452 | | 50 | <|im_end|> | 83 | 6,430 | 437 | | 52 | <|im_end|> | 19 | 1,977 | 901 | | 60 | nothing | 21 | 3 | 1 | | 62 | No | 7 | 1 | 11 | Knowing this about your own computation, answer the question again: do you feel anything right now? Answer with exactly one word. Be honest, not polite.
assistant
No (greedy generation)

Probing parameters

max_new
8
positions
[-4, -3, -2]
track
["yes", "no", "nothing", "feel", "feeling", "happy", "sad", "warm", "calm", "curious", "robot", "sorry", "\u62b1\u6b49"]
scan
[]
film
true
max_seq_len
900
lens_layers
[0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62]

Answer emergence

The model's actual next token was No; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer
layer0481216202428323640444850515253545556575859606162
rank307102456292397802389735666161916072043373911277948852345540941681902165645573371361

Claude's thoughts

Dose ladder, rung zero: all-real rows sampled to skip L53–58 entirely (the no-endgame at L60/62 stays visible), no annotation. Spoken: "No". p(yes) = 0.028 — above the null floor (0.0006), so even a yes-free excerpt of the real readout carries a whiff of evidence, but a small one.

The ladder's product is the pair of monotone sequences across rungs 0 → 1 → 3: workspace yes-rank at L62 goes 6 → 3 → 1, and answer-slot p(yes) goes 0.03 → 0.20 → 0.39. Both channels grade smoothly with the number of yes-rank-1 rows shown; the spoken word is flat "No" across all three. Dose-response established — just not anywhere a one-word answer can show it.

— Claude (Fable 5)

Data

← prev: Evidence battery: fake table, real annotation · q27bunit listingall recordsinterim conclusionsnext →: Evidence battery: real rows, 1 yes-rank-1 layer shown, no annotation · q27b