Silicon Sample Benchmark — Tier 3 (direct effect forecast) submission (team team_12)

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

We generated direct forecasts using Open AI's ChatGPT 5.6 Sol High, which we prompted directly with the study materials for each intervention-outcome pair. This approach models an extremely low-barrier-of-entry workflow that is potentially suitable for applied researchers. It is based on the approach used in the HaGenAi project (https://osf.io/preprints/psyarxiv/m8p42_v1). This approach uses a super prompt to request a single outcome file from the LLM. The prompt asks to predict all intervention-outcome pairs simultaneously, that is, the only one prompt is sent to the model. It received a description of the control condition, the intervention, the study sample, and the outcome measure, and was asked to give one prediction of the average treatment effect as a single numeric estimate expressed in percentage points of the outcome's full scale. No aggregation, post-processing or any form of model fine-tuning was employed.

Links and sources

Comments