engblogs

Examines OpenAI's public policy agenda, detailing frontier AI safety, governance and international standards, federal and state policy frameworks, transparency, and the drive to ensure AI benefits all of humanity and broad access.

Proposes a durable federal governance framework for frontier AI safety, integrating state-aligned standards, a central safety authority (CAISI), and a nationwide resilience plan to address national security and public safety risks as frontier AI advances.

A blueprint for democratic governance of frontier AI outlining how the U.S. can build a durable federal framework for frontier AI safety, harmonize state approaches, strengthen CAISI, and mobilize a national resilience plan to address national security and public safety challenges.

Wasmer leveraged Codex to build Edge.js, a Node.js runtime that runs in a WebAssembly sandbox at the edge, delivering full Node.js workloads without Docker in two weeks and achieving 10x–20x faster development.

Introducing an updated GPT-Rosalind that combines agentic coding and tool-use with stronger drug-discovery capabilities, broader life-sciences workflows, and the LifeSciBench benchmark to measure real-world impact at enterprise scale.

Wasmer leverages Codex to run Node.js workloads at the edge inside a WebAssembly sandbox, delivering Edge.js and a dramatically faster JavaScript runtime without Docker.

GPT-Rosalind expands enterprise-scale life-sciences AI with agentic coding and tool-use, stronger medicinal chemistry and genomics capabilities, LifeSciBench benchmarking, and plugins that connect literature, omics analyses, and wet-lab workflows.

Examines the rise of end-to-end programming benchmarks via ProgramBench, an artifact-focused, open leaderboard that tests frontier models on real software tasks with emphasis on rigor, reproducibility, human-in-the-loop interaction, and community-driven evaluation.

A concise look at ProgramBench and the future of coding benchmarks, focusing on artifact-level evaluation, rigorous verification, and the shift toward interactive, human-guided software construction.

MIT researchers present ChartNet, a synthetic-data pipeline that teaches vision-language models to robustly interpret charts, enabling open-source models to outperform larger commercial ones in chart reconstruction, data extraction, summarization, and question answering.

Tod Machover, MIT Media Lab director and pioneering figure in AI-driven music technology, receives the George Peabody Medal for groundbreaking work in participatory opera and the interdisciplinary fusion of music and technology.

MIT's ChartNet dataset trains open-source vision-language models to robustly interpret charts, enabling accurate data extraction, summarization, and chart-based reasoning at scale.