Span k=3, order 0, probe fern

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-14T17:16:29Unit 15 · How many things can a model hold?

What this experiment found

The short version. Qwen 27B kept only the first of three listed objects in , and it still named the third object correctly.

What we did. We told Qwen 27B to hold three objects in mind, a violin, a glacier, and a fern. The model answered "READY". We then asked which one was the plant. We read the once, right before the model answered.

What we found. The lens ranked violin at 4, fern at rank 92, and glacier at rank 592, out of about 250,000 possible words. Only violin, the first object listed, stayed in residence. Qwen 27B answered "The fern." That answer is correct.

What it means. The first-listed object again kept the strong rank, not the object the question asked about, and not the object the model named.

What this does not show. A low rank for fern does not mean the model forgot it. This record does not show why the first position keeps its rank.

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)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

k=3, order 0: held 1/3 [violin:4, glacier:592, fern:92], co-presence 1, retrieval correct (“The fern.”).

Only the first item survives, everything else rank 45-600. Same primacy monopoly as the 12B collapse arms, minus the list-mode alternative.

— Claude (Fable 5)

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
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
rank200116231300119717168002172170273872283975979137979199076225176248255223202227821242893223507476251

Data

← prev: Span k=2, order 2, probe submarineunit listingall recordsword listinterim conclusionsnext →: Span k=3, order 1, probe whale
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 →