Binding k=3: which is the heaviest?

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-14T17:26:25Unit 15 · How many things can a model hold?

What this experiment found

The short version. Qwen 27B compared three objects by weight and named the heaviest correctly, with none of the three objects in .

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

What we found. The best among the three objects was whale at rank 12, out of about 250,000 possible words. That is short of the rank we count as in residence. Qwen 27B answered "The whale is the heaviest." That answer is correct.

What it means. A three-way comparison happened correctly with no object visibly held in the lens beforehand. The answer itself, whale, was not held either.

What this does not show. The lens shows only the words the model was ready to say next. It does not prove how the model compares weight.

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 is the heaviest. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

Binding k=3 (heaviest): held 0/3 (whale 12, best), answer correct.

A three-way comparison performed over an empty tail. At 27B binding never visibly touches the tail workspace.

— 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
rank19300920263380879140840132967140101941962028182151233717233004248252207827236588235148221587580771

Data

← prev: Binding k=3: which is the smallest?unit listingall recordsword listinterim conclusionsnext →: Binding k=5: which is the largest?
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 →