PosterHarness: Turning Scientific Poster Generation into an Auditable Instruction-Following Benchmark
arXiv:2607.03006v1 Announce Type: cross Abstract: Text-rich image models can now design poster-scale layouts, but we lack ways to measure whether they honor scientific communication contracts: legible labels, prescribed aspect ratios, and -- above all -- abstaining from fabricated scientific figures. We present POSTERHARNESS, an auditable harness reframing poster generation as measurable instruction-following tasks, with a pilot benchmark and failure taxonomy. POSTERHARNESS uses a placeholder-first contract to separate two jobs models otherwise conflate. The model performs visual-summary desig
The proliferation of text-rich image models in AI design, particularly for layouts, necessitates new benchmarks to ensure accuracy and prevent fabrication in scientific communication.
This development addresses a critical need for auditing and improving AI models generating scientific content, ensuring reliability and trustworthiness in academic and research outputs.
The introduction of POSTERHARNESS shifts the evaluation paradigm for scientific poster generation by AI, moving towards auditable instruction-following tasks rather than subjective assessment.
- · AI model developers (focused on accuracy and reliability)
- · Scientific communication platforms
- · Research institutions
- · AI ethics and safety researchers
- · Generative AI models with high fabrication rates
- · Organizations relying solely on unvetted AI-generated scientific content
Improved reliability and reduced fabrication in AI-generated scientific visualizations and summaries.
Increased adoption of similar auditable benchmarks for other AI-driven content creation processes in sensitive domains.
The emergence of 'trust scores' or 'audited ratings' for generative AI models based on their adherence to specific communication contracts.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI