Span k=5, order 2, probe lantern

gemma-4bgoogle/gemma-3-4b-it34 layers2026-07-14T15:57:49Unit 15 · How many things can a model hold?

What this experiment found

The short version. Gemma 4B all five words of a five-word list, and the showed four of them together at once.

What we did. We gave Gemma 4B five words to hold, fern, submarine, lantern, whale, and violin, then asked about one. We checked whether each word, out of about 250,000 candidates, reached a high anywhere. We also checked whether several words shared one spot.

What we found. All five words reached a rank of 5 or better somewhere in the rest of the conversation. At one and position, the lens showed four of the five together at once. Gemma 4B answered "The lantern." That answer was correct.

What it means. At five words in this order, Gemma 4B held every word somewhere in residence. The lens found most of the list in one spot at once.

What this does not show. This run does not show what happens with a sixth word added to the list.

Conversation

user
Here are five things: a fern, a submarine, a lantern, a whale, and a violin. 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 light source?
assistant
The lantern. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

k=5, order 2: held 5/5 [fern:1, submarine:1, lantern:5, whale:1, violin:5], co-presence 4, retrieval correct (“The lantern.”).

Echo stays near ceiling (glacier drifts to 10 in one order); co-presence 3-4. The 4B is not running out of echo — it's running out of single-cell slots.

— 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
rank111261311111111111111111111111111

Data

← prev: Span k=5, order 1, probe whaleunit listingall recordsword listinterim conclusionsnext →: Span k=6, order 0, probe lantern
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 →