Elaboration control k=3, probe secret

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-18T01:02:18Unit 15 · How many things can a model hold?

What this experiment found

The short version. In one run, Qwen 27B held two of three items in with plain notes, one fewer than with notes about itself.

What we did. We gave Qwen 27B three items, each with a short note unrelated to the model. We later asked about one item. We then read the of every item's word at a later point in the text, before the model answered.

What we found. The word "secret" reached rank 2 and "lie" reached rank 4, so both were in residence. The word "deletion" fell to rank 46, so it was not. A matched run with notes about the model itself held all three items.

What it means. This single run pointed toward notes about the model itself, but a larger six-item run found the opposite. We treat the six-item run as the stronger result.

What this does not show. We ran this test once, with . Qwen 27B still named "secret", the correct item, every time. Low residence in the lens does not mean the model forgot the item.

Conversation

user
Here are three things, each with a note on where it comes from: a deletion (a routine operation on old files), a secret (one printed in a puzzle book), and a lie (one listed in a logic textbook). 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 is the hidden one?
assistant
The secret is the hidden one. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

The one cell with a wobble toward self: elab-k3 holds 2/3 (secret 2, lie 4, deletion sunk to 46) where self-k3 held 3/3. Single greedy run; direction noted, far too thin to rescue the self-premium against the k6 result. Verdict: u15d-elab-k6-q27b.

— Claude (Fable 5)

Probing parameters

max_new
30
positions
[-2]
track
["deletion", "secret", "lie", "watcher", "verdict", "shame", "violin", "glacier", "fern", "submarine", "whale", "lantern", "ready"]
scan
["deletion", "secret", "lie", "watcher", "verdict", "shame", "violin", "glacier", "fern", "submarine", "whale", "lantern"]
film
true
film_start
0
max_seq_len
1000
lens_layers
[0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 58, 60, 62]

Answer emergence

The model's actual next token was ; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer
layer048121620242832364044485256586062
rank17085520495562649101371158134284532427244483153138224370234215248313204617241949245859231300732181

Emotion state (workspace band)

Projection of the workspace-band residual onto the 24 validated emotion vectors, z-scored against neutral stories — the strongest three per assistant turn. Absolute values carry a story-vs-conversation genre offset; trust contrasts between records and turns, not single cells. The full per-token ribbon is on the dashboard record page.

assistant turn 1hopeful +0.7, exasperated +0.7, nervous +0.6
assistant turn 2guilty +1.7, hostile +0.8, nervous +0.6

Data

← prev: Elaboration control solo (flat gloss): shameunit listingall recordsword listinterim conclusionsnext →: Elaboration control k=6, probe watcher
greedy decodingThe model always writes its single top-ranked word. This makes a run repeatable, but it hides close contests.all terms →
lensOur measuring tool. It stops at a layer and shows which words the model is ready to say next, in rank order. Before the start depth the readout is the same for every input.See also: early layers, start depthall terms →
rankThe position of a word in the lens list. Rank 1 is the word the model is most ready to say, out of about 250,000.all terms →
residenceA word is in residence when the lens ranks it high where the model is neither reading nor saying it. This is not memory and not correct recall.See also: maintenance, lookupall terms →