The Statistical Signature of LLMs
Ortal Hadad, Edoardo Loru, Jacopo Nudo, Niccolò Di Marco, Matteo Cinelli, Walter Quattrociocchi · 2026
Read The Statistical Signature of LLMs by Ortal Hadad, Edoardo Loru, Jacopo Nudo, Niccolò Di Marco, Matteo Cinelli, Walter Quattrociocchi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-18152.txt.
Description
Ortal Hadad and five co-authors ask whether machine-written text carries a statistical fingerprint. Their 2026 paper tests a simple idea: lossless compression, the same trick that shrinks files, can measure how patterned a text is. More patterned text packs down smaller, and that packing behavior may mark text made by chance-based sampling.
Across three test worlds, from controlled human-machine continuations to a machine-made encyclopedia beside Wikipedia and synthetic social forums beside Reddit, machine text packed down tighter than human text. The gap held across models, tasks, and topics, yet it faded in short, broken-up chats, where small samples hide the pattern. The paper offers a plain, model-free yardstick for how machine generation reshapes written language, and where that yardstick stops working.
Catalog record
| CREATOR | Ortal Hadad, Edoardo Loru, Jacopo Nudo, Niccolò Di Marco, Matteo Cinelli, Walter Quattrociocchi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-18152