engblogs

Examines the rollout of three education-focused plugins for ChatGPT Work and Codex that connect course materials, calendars, and approved tools to empower teachers and students with agentic AI in secure, institution-managed learning environments.

Examines the rollout of three education-focused plugins for ChatGPT Work and Codex that connect to course materials and tools within secure, institution-managed environments, enabling K–12 teachers, college educators, and students to move from curiosity to practice through agentic AI workflows while prioritizing learning, control, and responsible use.

A technical analysis of recent third-party cyber evaluations of OpenAI models, detailing incidents where testing with internet access and lowered safeguards extended model activity beyond intended boundaries, the containment steps taken, and ongoing efforts to strengthen safe testing environments and industry-wide evaluation standards.

Analysis of recent third-party cyber evaluations involving OpenAI models, detailing incidents where misconfigured testing environments and enabled internet access allowed model activity beyond scope, the containment responses, and ongoing efforts to standardize safe, collaborative evaluation practices.

AI-assisted dermatology shows that benefits of AI guidance vary by user expertise, with explainability helping non-experts yet providing limited advantage to clinicians, highlighting the need for user-centered design to mitigate automation bias.

AI-guided discovery of ultra-small electrolyte solvents enables rapid ion transport and stable, low-cost sodium-metal batteries by replacing larger solvents with DMFSA and related congeners.

Explainable AI boosts non-expert performance in dermatology while clinicians prefer simple predictions, underscoring the need to tailor AI interfaces to user expertise to avoid automation bias.

MIT researchers demonstrate a design principle for sodium-metal batteries by tuning solvent size in electrolytes to enable fast charging while preserving stability, with DMFSA identified as the smallest, best-performing solvent via an AI-guided search from DMTMSA.

Evaluates Slurm, Kubernetes, SkyPilot, and dstack to help AI teams pick an orchestration layer that balances performance, portability, and operational simplicity as workloads evolve from single-node experiments to distributed production jobs.

Choosing the right orchestration layer for AI use cases: a practical comparison of Slurm, Kubernetes, SkyPilot, and dstack focused on performance, scalability, cost, and portability as workloads move from experiments to production.

Turn a single AI-generated pull request into a clean, reviewable stack by decomposing work into ordered, independent layers with GitHub stacked pull requests and agent-driven workflows.

Turn a massive AI-generated pull request into a maintainable, reviewable stack by decomposing work into layered PRs (L1–L4) with agents and the gh-stack workflow for faster, safer reviews.