Span k=2, order 0, probe glacier

gemma-4bgoogle/gemma-3-4b-it34 layers2026-07-14T15:43:44Unit 15 · How many things can a model hold?

What this experiment found

The short version. Gemma 4B two words, violin and glacier, both at in the , and named the right one when asked.

What we did. We gave Gemma 4B a list of two words, violin then glacier, and later asked about one of them. We read the rank of each word, out of about 250,000 candidates. We also checked whether both words showed up together at the same and position.

What we found. Both violin and glacier held rank 1 through the rest of the conversation. The lens showed both words together in the same spot, a of two out of two. Gemma 4B answered "The glacier." That answer was correct.

What it means. Two words is an easy case for Gemma 4B at this size. Both stayed in residence at the same time, and the spoken answer matched the lens.

What this does not show. This run does not show what happens with a longer list. Other runs in this unit raise the number of words.

Conversation

user
Here are two things: a violin and a glacier. 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 ice?
assistant
The glacier. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

k=2, order 0: held 2/2 [violin:1, glacier:1], co-presence 2, retrieval correct (“The glacier.”).

Both items rank 1 through the tail, co-present at a single position. Trivial for the 4B.

— 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

Answer emergence

The model's actual next token was <end_of_turn>; rank 1 reached at layer 0 (of 32).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132
rank111151411111111111111111111111111

Data

← prev: Span solo baseline: lanternunit listingall recordsword listinterim conclusionsnext →: Span k=2, order 1, probe whale
co-presenceThe number of tracked words that the lens ranks high in the same place at the same depth.all terms →
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 →