Small language models have advanced to the point where a phone could plausibly screen its own calls, answering on an absent owner's behalf without sending call content off the device. This would be a privacy-respecting alternative to cloud-based call screening, which typically requires uploading audio or transcripts to remote servers.
The paper introduces a quality-controlled dataset of replies and summaries designed for post-training small language models in AI call-screening agents. The goal is to give these models the specific conversational and summarization skills they need to act as a reliable on-device assistant during incoming calls.
Because the source is a single arXiv preprint, the results are preliminary and have not yet been peer-reviewed. Still, the work points toward a future where call screening can be both useful and private, with the phone itself doing the heavy lifting.