engblogs

Researchers leverage AlphaFold and molecular simulations to engineer heat-stable photosynthesis enzymes, paving the way for crops resilient to global warming.

An in-depth look at the OpenAI Foundation's People-First AI Fund: 208 unrestricted grants to community-based nonprofits nationwide, advancing AI literacy, local impact, and equitable access.

The OpenAI Foundation announces $40.5 million in unrestricted grants to 208 community-based nonprofits nationwide through the People-First AI Fund, supporting AI literacy, economic opportunity, and transformative community impact.

Trains language models to explicitly confess when they break instructions, improving visibility of misbehavior to enhance monitoring, training, and safety.

Introducing a novel 'confessions' technique that trains language models to self-report misbehaviors and increase transparency by separating honesty-focused outputs from primary responses.

OpenAI is acquiring neptune.ai to enhance AI model training by integrating advanced experiment tracking and analysis tools for improved research and development.

Pinterest's observability team leverages AI-driven Model Context Protocol (MCP) to unify fragmented telemetry data, enabling autonomous root-cause analysis and enhanced engineering insights without overhauling existing infrastructure.

Macro, an advanced open-source modeling tool developed by MITEI and collaborators, enables power-system planners and policymakers to simulate and optimize future energy infrastructure and policies in real time, addressing complex interdependencies across sectors to ensure reliable, low-cost, and low-emission electricity systems.

MIT engineers developed an AI-powered aerial microrobot with insect-like speed and agility, capable of executing complex maneuvers such as somersaults using a novel two-step control system combining model-predictive control and imitation learning.

Agentforce enhanced enterprise AI response times by 3–5x through a restructured architecture separating deterministic logic from LLM reasoning, consolidated data retrieval, and a scalable multi-brand conversational AI design.

Exploring the evolution of testing strategies at Jane Street, highlighting the use of Antithesis for end-to-end automated chaos testing to uncover elusive bugs and achieve battle-tested reliability in distributed systems like Aria.

Cloudflare's Q3 2025 DDoS Threat Report reveals unprecedented hyper-volumetric attacks by the Aisuru botnet, surging threats targeting AI and critical industries, and highlights the rising complexity and scale of global DDoS attacks amidst geopolitical tensions.