Show Me Science

Key Takeaways

• LLMs may become primary consumers of research, reducing narrative need
• Authors often omit task examples to squeeze more results into limited space
• Showing plots instead of tables highlights effect size and uncertainty
• Providing prompts, interfaces, and failure cases improves reproducibility and reviewer judgment
• Community guidelines like “show me the raw output” can curb speculative spin

Pulse Analysis

The rise of large language models as de‑facto reviewers is prompting a reassessment of how scientific findings are packaged. Historically, scholars have debated trimming discussion sections to curb spin, but the advent of AI‑driven consumption accelerates the push toward concise, data‑first formats. Journals and pre‑print servers are now grappling with whether narrative prose is essential when algorithms can parse tables, code, and raw outputs directly, potentially rendering traditional storytelling obsolete.

Advocates of the "show me" movement argue that concrete artifacts—plots, variance visualizations, interface screenshots, prompts, and failure cases—provide the evidentiary backbone reviewers need. By replacing dense tables with intuitive graphs, readers can instantly gauge effect sizes and uncertainty, reducing misinterpretation. Publishing the exact prompts used in LLM experiments or the raw qualitative codes for model outputs equips peers to replicate studies and spot hidden biases. Such transparency not only strengthens reproducibility but also curtails the temptation to inflate claims through rhetorical flourish.

If the community embraces these practices, the next generation of scientific communication could resemble a structured data repository rather than a narrative essay. Publishers may adopt new submission templates that allocate space for visual and raw artifacts, while funding agencies could tie grants to open‑access of these materials. Balancing the need for human‑readable context with algorithm‑friendly detail will be crucial; narrative will likely persist as a guide, but the core of scientific validation will shift toward demonstrable, reproducible evidence.