Samuele Sala

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 2,500 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions.

We present compelling evidence for the existence of an evolutionary adaptive response to viral agents such as SARS-CoV-2, that results in the human in vivo biosynthesis of a family of compounds with potential antiviral activity. Using nuclear magnetic resonance (NMR) spectroscopy, we detected a characteristic spin-system motif indicative of the presence of an extended panel of urinary and serum metabolites during the acute viral phase. The structure of eight of nucleoside analogues was elucidated (six of which have not previously been reported in human urine), and subsequently confirmed by total-synthesis and matrix spiking. The molecular structures of the nucleoside analogues and their correlation with an array of serum cytokines, including IFN-α2, IFN-γ and IL-10, suggest an association with the viperin enzyme contributing to an endogenous innate immune defense mechanism against viral infection.