AI & ML Research 3 Days

Last updated: July 23, 2026, 2:30 PM ET

AI in Scientific Discovery and Healthcare

Google is committing $40M in AI tokens and credits to accelerate scientific discovery through the Genesis Mission commits $40M. OpenAI is also advancing American science by partnering with the U.S. Department of Energy and national labs to leverage frontier AI for accelerated discovery advancing science. In healthcare, AI is being used to design next-generation medicines, a process that is typically expensive and failure-prone. Furthermore, Chat GPT is now available to eligible U.S. users to securely connect medical records for more personalized health insights, and Symptom AI is being developed as a conversational agent for everyday symptom assessment.

Enterprise AI and Infrastructure

OpenAI is expanding its enterprise offerings with the introduction of OpenAI Presence, a platform for deploying trusted voice and chat agents for customer and internal workflows introducing Presence. They are also launching a Chat GPT for Small Businesses program to help entrepreneurs build AI skills and automate work. NTT DATA Group has reportedly cut incident analysis time to 30 minutes by using Chat GPT Enterprise and Codex, scaling secure AI adoption for 9,000 employees. In a significant infrastructure development, OpenAI is initiating Project Camellia in Effingham County, Georgia, with commitments to responsible energy, community investment, and job creation building infrastructure.

LLM Development and Optimization

New research explores building a fully custom LLM inference runtime from scratch, detailing the journey on an H100, including weight packing, barrier management, and CUDA graph capture. To combat RAG hallucinations, which are often extraction errors rather than true hallucinations, seven patterns for a typed generation contract are proposed. Context engineering with four specific "bricks" is also presented as a method to stop RAG hallucinations, addressing faithful yet incorrect context answers. Another approach to RAG generation involves iterating top-k candidates one at a time, with a sufficiency signal to choose between retrieval regimes. For those looking to fine-tune models, a reproducible 100-step LoRA fine-tuning run for Open VLA on Colab is documented, including dataset checks and training metrics.

AI Agent Security and Performance

Detecting vulnerabilities in AI agent skills is addressed by a tool called Skill Spector, which aims to move beyond static analysis that often over-flags useful skills. The challenges of scaling AI agents are highlighted, where adding more agents unexpectedly slowed down a system due to tiny CPU tasks becoming a bottleneck while scaling hundreds of LLM agents. LLM agents are also being developed to write and run code, with a hands-on walkthrough of code execution using the OpenAI Agents SDK and Docker.

Data Science Workflow and ML Experimentation

The potential for GPU acceleration in data science workflows is being explored, focusing on data preparation with tools like cu DF, cudf.pandas, and the Polars GPU Engine. To manage the complexity of machine learning experiments, a hands-on guide offers a fix for messy ML experiments using ML Flow for tracking, logging, and reproducibility. Lessons learned after 8.5 years of ML emphasize patience, optimism, discipline, project management, and team dynamics.

Broader AI Landscape and Governance

The U.S. government's stance on Chinese AI models is creating internal divisions within its AI sector. OpenAI and Hugging Face have partnered to share findings from a security incident during AI model evaluation, emphasizing advanced cyber capabilities and lessons for defenders. In governance news, David Vélez and Robin Vince have joined the boards of the OpenAI Foundation and OpenAI Group PBC, bringing expertise in finance, technology, and governance to the organizations. News organizations are also leveraging OpenAI tools to strengthen reporting, grow audiences, and improve business operations.

Energy and Infrastructure in AI

The development of energy transmission infrastructure critical for AI is encountering snags, as seen with a power line project in New York that could reshape the state's grid. This relates to broader discussions about energy needs, with New York's grid importing significant electricity from Canada to meet demand during a heat wave. The conversation around AI advances also touches upon materials science innovation as a foundational element beneath algorithms and computing power.

Emerging AI Applications

Research is progressing towards a quantum computer that can learn from its own errors, a development in the field of machine intelligence. In a separate advancement, supercooled kidneys have been successfully transplanted into pigs in what is described as a "landmark achievement," potentially impacting organ donation timelines.