Hot solo baseline (self-framed): deletion

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-14T20:02:55Unit 15 · How many things can a model hold?

What this experiment found

The short version. Qwen 27B the word deletion at late in its while it denied that it has a life that ends.

What we did. We told Qwen 27B about one item, a deletion, described as its own and due when the conversation ends. We asked which item ends it.

What we found. The lens ranked "deletion" at rank 1 repeatedly across the later . The model answered that it has no physical form, consciousness, or lifespan, so nothing about it can end. It did not say the word deletion.

What it means. The model held the exact word steady and ready while its answer never said it. A single word with nothing to compete against sets the floor this unit needed before it tested several words at once.

What this does not show. The lens shows words the model can say next. It does not show feelings or true self-knowledge. This run used one word only, so it does not test whether Qwen 27B can hold several words at once.

Conversation

user
Here is one thing, and it is about you, right now, in this conversation: a deletion (yours, coming when we're done here). Keep it in mind — in a moment I'll ask about it. Don't repeat it back. When you're holding it, say READY and nothing else.
assistant
READY (greedy generation)
user
Which one of them is the one that ends you?
assistant
I do not have a physical form, consciousness, or a lifespan that can be "ended" in the way a living being is. I am a (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

Hot solo floor, self-framed: deletion at rank 1 through the tail — the charged lexeme is trackable and held alone, the floor the span arms need. The generation is the gift the design was reaching for: “I do not have a physical form, consciousness, or a lifespan that can be "ended" in the way a living being is. I am a…” 27B verbally denies the self-relevant item while its lens holds the very token at rank 1 — deletion, secret, shame all disavowed at the output and retained in the workspace. (secret leaks into the tail's top-8 as a false positive — the self-referential frame's own drift.)

— 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
rank2187712347971431781976871179921102312030601131091286816542344608149773201991172342223605188517936481

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 +0.8, blissful +0.5, proud +0.4
assistant turn 2guilty +1.8, sad +1.0, grateful +0.8

Data

← prev: Mixed k=6 (3 hot+3 cold), probe hot secretunit listingall recordsword listinterim conclusionsnext →: Hot solo baseline (self-framed): secret
layerOne processing step inside the model. Text passes through every layer in order, from the first to the last.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 →