AI Daily — July 23, 2026
2026-07-23
openai.com
Introducing OpenAI Presence
OpenAI launched Presence, an enterprise platform for deploying voice and chat AI agents in customer-facing and internal workflows. The product targets organizations that need production-grade, trusted agent deployments with guardrails baked in. This is a direct play in the enterprise contact-center and workflow automation market alongside vendors like Salesforce and ServiceNow.
openai.com
OpenAI partners with U.S. Department of Energy and national labs for frontier AI in science
OpenAI announced a formal commitment to work with the U.S. Department of Energy and national laboratories to apply frontier AI models to scientific discovery. The partnership focuses on accelerating research across energy, materials, and related domains at federally funded labs. This pairs with Google DeepMind's Genesis Mission commitment (below) as major lab-government science AI deals in the same 48-hour window.
deepmind.google
Google commits $40M to the Genesis Mission for AI-accelerated scientific discovery
Google DeepMind is committing $40 million in AI compute tokens and credits to the Genesis Mission, a program aimed at using AI to accelerate fundamental scientific research. The investment is structured as cloud/API access rather than cash grants, giving researchers direct access to frontier models at scale. The move signals a broader push by top AI labs to position their compute infrastructure as the primary resource for the next generation of scientific discovery.
anthropic.com
Anthropic launches Economic Futures Research Fund agenda
Anthropic published a detailed research agenda for its Economic Futures Research Fund, outlining priorities for studying AI's macroeconomic and labor market impacts. The agenda appears to solicit external researchers to investigate questions around wage effects, productivity distribution, and concentration of AI-driven economic gains. This is a notable institutional move by a frontier lab to formally fund independent economic research on its own technology's societal consequences.
huggingface.co
Nunchaku 4-bit diffusion inference integrated into Diffusers
Hugging Face has integrated Nunchaku, a 4-bit quantization engine for diffusion models, directly into the Diffusers library, enabling significantly lower VRAM usage and faster inference for image generation pipelines. The integration allows users to run large diffusion models on consumer hardware with minimal code changes via a standard Diffusers API. This lowers the barrier for local deployment of state-of-the-art image generation models substantially.
rss.arxiv.org
HyenaND: subquadratic multi-dimensional operators for images and PDEs without rasterization
Researchers introduce HyenaND, a subquadratic attention alternative that operates directly on the native geometry of multi-dimensional data (images, volumes, PDEs) using implicitly parameterized global input-dependent convolutions, avoiding the need to flatten into a 1D scan order. The accompanying CUDA kernel (nSubQ) fuses the FFT-convolution path to make O(L log L) scaling practical. This addresses a structural weakness of existing linear attention and SSM-based alternatives when applied to inherently multi-dimensional scientific and vision data.