POaaS: Minimal-Edit Prompt Optimization as a Service to Lift Accuracy and Cut Hallucinations on On-Device sLLMs
Jungwoo Shim, Dae Won Kim, Sun Wook Kim, Soo Young Kim, Jae-geun Cha, Hyunhwa Choi · 2026
Read POaaS: Minimal-Edit Prompt Optimization as a Service to Lift Accuracy and Cut Hallucinations on On-Device sLLMs by Jungwoo Shim, Dae Won Kim, Sun Wook Kim, Soo Young Kim, Jae-geun Cha, Hyunhwa Choi (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-2603-16045.txt.
Description
Shim and co-authors polish user prompts on the device itself to lift small model accuracy. Phone-grade models stumble on typos, vague asks, and missing context. Cloud prompt optimizers write long rule sheets that waste small context windows. Their POaaS layer routes each query to light specialist editors. A Cleaner, a Paraphraser, and a Fact-Adder each make minimal edits. Strict drift and length caps merge the edits safely. Clean prompts skip the layer under a cautious rule. Tests on two Llama models show accuracy and factuality gains.
Catalog record
| CREATOR | Jungwoo Shim, Dae Won Kim, Sun Wook Kim, Soo Young Kim, Jae-geun Cha, Hyunhwa Choi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49270 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.16045