AI & ML Research 3 Days

Last updated: July 24, 2026, 2:30 PM ET

*LLM Advancements and Applications

New foundation models are emerging for tabular data, achieving performance surpassing traditional gradient-boosted trees on the Tab Arena benchmark by predicting missing spreadsheet columns zero-shot. For retrieval-augmented generation (RAG) systems, researchers are exploring loop engineering techniques to create LLM cascades, starting with local models and escalating to more powerful hosted ones, incorporating validation loops and per-question type dispatchers to manage costs and improve accuracy. A key insight into RAG errors suggests that most "hallucinations" are actually extraction errors, stemming from the model's reading of the context, which can be addressed by implementing typed generation contracts. To improve AI agent memory, a usage-reinforced decay engine based on the Ebbinghaus forgetting curve is proposed, prioritizing important information over simply the newest data. Building custom LLM inference runtimes from scratch, including managing weights and CUDA graphs on hardware like the H100, is now a documented journey for those seeking deep control.

AI in Enterprise and Operations

OpenAI is launching OpenAI Presence, an enterprise AI agent platform designed for deploying voice and chat agents in customer and internal workflows. News organizations are leveraging OpenAI tools to enhance reporting, expand their reach, and streamline business operations. NTT DATA Group has significantly reduced incident analysis time to just 30 minutes by utilizing Chat GPT Enterprise and Codex, enabling 9,000 employees to automate tasks and scale secure AI adoption. In a move towards responsible AI development and community investment, OpenAI has announced Project Camellia in Effingham County, Georgia, focusing on responsible energy use, job creation, and access to Codex.

AI for Scientific Discovery and Health

Google is committing $40 million in AI tokens and credits to support the Genesis Mission, aiming to accelerate scientific discovery committing AI resources. In the health sector, Chat GPT is introducing a "Health in Chat GPT" feature for eligible U.S. users, allowing secure connection of medical records and Apple Health data for personalized health insights. Google AI is also developing Symptom AI, a conversational AI agent designed for everyday symptom assessment. Furthermore, research is progressing towards quantum computers that can learn from their errors, a significant step in machine intelligence. AI is also proving instrumental in designing the next generation of medicines, a process that is traditionally expensive and prone to failure.

Infrastructure, Agents, and Security

Efforts are underway to build intelligent document processing (IDP) systems in the cloud, automating the classification and extraction of personally identifiable information (PII) from emails, particularly using AWS services. The integration of multiple AI agents into a system can paradoxically lead to performance degradation, with tiny CPU tasks becoming a significant bottleneck when scaling hundreds of LLM agents asynchronously. For AI agents, detecting vulnerabilities in their skills is crucial; static analysis tools can identify malicious skills but also over-flag useful ones, highlighting the continued need for human judgment in security assessments. Building an LLM agent capable of writing and executing code is now achievable through practical walkthroughs using the OpenAI Agents SDK and Docker.

*Broader Technology Trends and Learnings

In a significant development, supercooled kidneys have been successfully transplanted into pigs in what is described as a "landmark achievement," offering potential breakthroughs in organ donation and transplantation. This achievement is part of a broader discussion on technological advancements, including homegrown Chinese chips and the potential reshaping of New York's energy grid by new power lines. From a personal perspective, an engineer shares lessons learned after 8.5 years in machine learning, emphasizing the importance of patience, optimism, discipline, project focus, and teamwork. The complexities of data science are also highlighted, with a story illustrating a scenario where data-driven decisions led to an undesirable outcome, underscoring the human element in analytics.