AI & ML Research 3 Days

Last updated: July 22, 2026, 11:30 PM ET

AI Development & Infrastructure

OpenAI announced Project Camellia in Effingham County, Georgia, committing to responsible energy, community investment, job creation, and access to Codex. The company also launched the Chat GPT for Small Businesses program to help entrepreneurs build AI skills and automate work. NTT DATA Group is leveraging Chat GPT Enterprise and Codex to automate tasks for 9,000 employees, reducing incident analysis time to 30 minutes and scaling secure AI adoption. OpenAI is also partnering with Hugging Face to address a security incident during AI model evaluation, sharing early findings and lessons for defenders. Furthermore, David Vélez and Robin Vince have joined the boards of the OpenAI Foundation and OpenAI Group PBC, bringing expertise in finance, technology, and governance.

Google AI is committing $40M in AI tokens and credits to the Genesis Mission to accelerate scientific discovery Google commits $40M to Genesis Mission. The company is also exploring a quantum computer that can learn from its errors. Meanwhile, MIT Technology Review notes that beneath AI advancements often lie innovations in materials science.

LLM Engineering & Applications

Towards Data Science offers a guide on building an LLM inference runtime from scratch on an H100, covering weight packing, CUDA graphs, and all barriers. For Retrieval Augmented Generation (RAG), "Loop Engineering for RAG Generation" suggests iterating top-k candidates one at a time and details two regimes for sending retrieved candidates to the generation brick, including a sufficiency signal for selection. Another article from the same publication argues that RAG systems aren't hallucinating but rather faithfully answering the wrong context, proposing "four bricks of context engineering" to mitigate this issue on real-world documents.

Engineers can explore running Claude code agents for extended periods, over 24 hours, to enhance productivity. For those looking to build LLM agents that can write and execute code, a hands-on walkthrough using the OpenAI Agents SDK and Docker is available. Security is also a concern, with "Detecting Vulnerabilities in Agent Skills with Skill Spector" highlighting how static analysis can miss malicious skills while over-flagging useful ones, emphasizing the role of human judgment.

OpenAI is introducing OpenAI Presence, an enterprise AI agent platform for deploying trusted voice and chat agents for customer and internal workflows. The company is also outlining its commitment to advancing American science by working with the U.S. Department of Energy and national labs to use frontier AI for accelerating discovery. OpenAI is sharing lessons learned from deploying long-running AI models, addressing new safety risks, observed failures, and improved safeguards through iterative deployment.

Data Science & ML Operations

For managing ML experiments, a hands-on guide provides fixes for messy experiments, covering tracking, logging models, and reproducing results with ML Flow. A reproducible 100-step LoRA fine-tuning run for Open VLA is detailed, including dataset checks, Colab setup, training metrics, and Weights & Biases evidence. Exploring GPU acceleration for data science workflows, one article focuses on accelerating data preparation using cu DF, cudf.pandas, and the Polars GPU Engine.

In the realm of enterprise document intelligence, "Loop Engineering with Adaptive Parsing" demonstrates parsing flat tables with Azure and figures with a vision LLM, positioning the LLM as a last line of defense. Another article discusses automatically assigning categories to uncategorized rows in Power Query and DAX based on specific rules, crucial for reporting and aggregation.

AI Ethics & Societal Impact

Researchers are examining AI's role in hiring, with findings suggesting AI is more likely than humans to form biases when screening résumés. News organizations are increasingly using AI tools to strengthen reporting, grow audiences, and improve business operations. Google AI is developing "Symptom AI," a conversational AI agent for everyday symptom assessment. In broader tech news, MIT Technology Review reported on China's AI models potentially dividing the AI world within the Trump administration. Discussions on AI also touch upon fundamental concepts like Byzantine Fault Tolerance, exploring decision-making in untrusted environments.