Paper2Agent is an AI agent that reads scientific papers and automatically reproduces their results. It's published in Nature and available as a live demo where you can query it about papers and run workflows through GitHub.

• Automates the extraction and reproduction of experimental results directly from published papers
• Reduces manual work scientists spend reverse-engineering methods and validating findings
• Deployed as an interactive agent you can query in real-time about paper contents and methodology

This paper argues that traditional academic articles hide failed experiments and leave out key implementation details, creating a “narrative tax” and an “engineering tax” that limit reproducibility. It proposes replacing static papers with ARA research packages—complete, executable bundles containing code, pipelines, and failure logs—so AI agents can fully understand and build on the work.

• 37 researchers from Stanford, CMU, Michigan and other top schools are pushing to replace traditional papers with "AI-ready research packages" (ARA)
• Papers hide a "narrative tax" (failed experiments and dead ends erased for a clean success story) and an "engineering tax" (missing implementation details AI needs to reproduce results)
• ARA packages would include full datasets, executable pipelines, complete code, decision logs, and records of failed attempts, not just a polished writeup
• Proposes judging research impact by whether an AI can rerun and build on it, rather than by narrative quality or page count

This article digs into why repeated LLM calls can produce different outputs even at zero temperature. It shows that floating-point non-associativity and kernel implementation details—rather than thread scheduling or atomic adds—are the real sources of run-to-run variation and outlines ways to make inference fully reproducible.

• Bit-identical results across thousands of runs of the same GPU matmul disprove the standard "concurrency + floating-point non-associativity" explanation for LLM nondeterminism.
• The actual cause is that libraries like cuBLAS/cuDNN nondeterministically pick among multiple reduction/tiling algorithms at runtime, each summing floats in a different order.
• Forcing a single deterministic algorithm choice (via flags like CUBLAS_WORKSPACE_CONFIG, torch.use_deterministic_algorithms(True), and seeded CUDA RNG) makes LLM inference output bit-for-bit reproducible.