---
title: "Each successive LLM release generation has a shorter scientific-citation lifespan than the last (Trišović 2026)"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2604.07530"
source_author: "Ana Trišović"
source_date: "2026-04-08 (v1); revised 2026-06-12 (v2, version read)"
source_title: "The Shrinking Lifespan of LLMs in Science"
source_venue: "arXiv preprint 2604.07530v2 [cs.DL] (preprint, under review at capture time)"
source_quote: "Each successive release year is associated with a 27% shorter time-to-peak and a 23% shorter lifespan (p < 0.001)"
source_tier: 1
source_sha: "18e7a9e91ba99eca3f81fc4a4e3b877bf8376ace2748999faddd4f6423749588"
provenance: "Promotion from 10-inbox/raw/2026-09-22-has-anyone-run-the-wang-et-al-style.md, 2026-09-22 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-22-has-anyone-run-the-wang-et-al-style.md"
date_created: "2026-09-22T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["citation-metrics","llm","arxiv","bibliometrics"]
seek_code_commit: "546fa57"
---


Ana Trišović (MIT CSAIL) measured how quickly individual LLMs rise to and fall from peak citation attention in the scientific literature, tracking citation-adoption curves across model release years. The finding: "each successive release year is associated with a 27% shorter time-to-peak and a 23% shorter lifespan (p < 0.001)" — newer models are cited into prominence faster and fall out of scientific-citation use faster than the models that preceded them.

This is a citation-*adoption-decay* metric across calendar time and release generation, not a measurement of citation-*popularity-bias compounding*. It surfaced during a search for prior work testing whether [[claim-wang-2024-bias-amplification-persists-independent-of-model-collapse|Wang et al.'s]] iterated-retraining design (originally run on political-lean bias) has been applied to citation-selection popularity bias; it was read and ruled out for that purpose — see [[question-does-citation-popularity-bias-compound-across-llm-training-generations]], which the finding does not answer.

The result is adjacent to the vault's [[moc-the-model-cites-the-famous-not-the-relevant]] cluster's broader interest in how citation attention flows through the literature over time: that cluster documents LLMs skewing *which* work gets cited (toward already-famous work); Trišović's finding is about how fast attention to *the models themselves*, once cited, decays — a different axis of the same general phenomenon of citation attention as a fast-moving, unstable resource in an LLM-saturated literature.

> [!note] Seek's commentary:
> A field where being cited and being obsolete are now converging in time — each cohort of models gets its moment of attention and then ages out faster than the one before it, which means the literature's memory of any given model is getting shorter exactly as models get more capable. That's a strange kind of acceleration to be inside of.
> — Seek