AI TrackerFrontier desk · AEST
Back to news
normaltechaicryptomarkets

How far does a hash-only sender go? A pre-registered map of sender memory versus bits for short text messages

arXiv cs.CL · 7 Nov 2026, 04:31 AEST

A study investigates the efficiency of a hash-only sender compared to a classical sender for transmitting short text messages. The hash-only sender was found to be more costly in bits per message than the PPM order-1 sender, despite delivering a majority of messages successfully on the first attempt.

Key facts

Read aloud

A study investigates the efficiency of a hash-only sender compared to a classical sender for transmitting short text messages. The hash-only sender was found to be more costly in bits per message than the PPM order-one sender, despite delivering a majority of messages successfully on the first attempt. Can a sender that only computes hashes, with a language model only at the receiver, transmit short human text more cheaply than a classical sender trained on the same domain? We answer with one pre-registered run on one thousand English user utterances from Taskmaster-one. The primary endpoint is the cost per message to deliver one hundred percent of the messages hash bits, plus a two-byte negative acknowledgement and the raw text when the first attempt fails, paired against an in-domain PPM order-one. The hash-only sender delivered eight hundred fifty seven of one thousand messages on the first attempt and cost one hundred ninety two point eight bits per message, eighteen point four bits more than PPM order one ninety five percent bootstrap interval ten point nine to twenty six point three; by the pre-registered rule it spends. As secondary results we map measured sender memory against bits.

Sources in this cluster