Direct Preference Optimization Moves Beyond Chatbots
A Hugging Face blog post explains how DPO, a simpler alternative to RLHF, can align models for tasks like summarization and translation.
AI & ML
Frontier models, research results, agents, and developer tooling.
Google Open-Sources Harness2 to Improve AI Agents at Test Time
The Apache-2.0 framework rewrites an agent's harness during inference, yielding an 11.5-point gain on WorkBuddy-Bench without retraining the underlying model.
Codex expands beyond coding into general knowledge work
OpenAI is positioning Codex as an everyday productivity tool for analysts, marketers, and other non-programmers, backed by a new research report.
Travelers rolls out OpenAI-powered claims assistant nationwide
The insurer says the AI tool will guide customers through claims and provide round-the-clock support, with a focus on scaling during peak periods.
OpenAI Clarifies Stance on AI Policy and Political Advocacy
OpenAI's new policy statement outlines its commitment to transparent, safety-focused regulation and distances the company from external political groups.
OpenAI Proposes Federal Framework for Frontier AI Safety
OpenAI's new blueprint outlines a U.S. federal approach to governing advanced AI, focusing on safety, resilience, and national security.
Enterprise AI Scale Needs Agent Logic, Not Just Larger LLMs
IBM Research argues that agent logic—not model size—is the missing layer for dependable enterprise AI adoption.
TinyFish Drops Search Fees, but Token Costs Remain
Free search APIs don't eliminate the real expense of AI agents: processing retrieved content.
Holo3.1 Brings Fast, Local Computer-Use Agents to Hugging Face
Hugging Face introduces Holo3.1, a computer-use agent system focused on speed and on-device execution.
Google Research: smartphone camera could enable passive heart-health monitoring
Google researchers describe a smartphone-camera technique that could turn everyday device use into passive heart-health tracking.
Endava Adopts OpenAI Agents to Accelerate Software Delivery
The IT services firm is integrating AI agents, ChatGPT Enterprise, and Codex into its workflows, signaling a shift toward AI-native engineering.
OpenAI Sets Out AI-First Biodefense Action Plan
A new OpenAI paper frames AI-powered biological resilience as a strategic priority for the Intelligence Age.
NVIDIA Nemotron 3.5 Offers Customizable Multimodal Content Safety
A new safety model from NVIDIA aims to give enterprises flexible, multimodal guardrails for AI deployments across global markets.
OpenAI adds ChatGPT memory system to track user preferences
A new memory capability aims to keep context fresh and relevant across conversations, making ChatGPT more helpful over time.
Voxel51 3.0 Unifies Multimodal Data Workflows for Physical AI
Voxel51 3.0 combines search, curation, annotation, and evaluation in a single multimodal data platform, adding an agentic toolset and a free tier.
OpenAI Proposes Industrial Policy for AI Era
OpenAI outlines a people-first framework for managing advanced intelligence through expanded opportunity and resilient institutions.
OpenAI Policy Push: Youth Safety and Global AI Standards
OpenAI's new policy agenda and youth-safety proposal signal a shift toward shaping AI governance internationally, with an emphasis on protecting young users.
OpenAI Adds Life-Science Capabilities to GPT-Rosalind
GPT-Rosalind expands into medicinal chemistry, genomics, and experimental workflows for research use.