AI & ML Research 24 Hours

Last updated: September 22, 2026, 10:05 PM ET

OpenAI Ships GPT-6 Sol and Luna

OpenAI introduced GPT-6 Sol and Luna, a pair of models positioned as bringing frontier intelligence to everyday use. The dual-model release splits the lineup between two capability tiers, continuing the company's pattern of pairing a flagship with a lighter, faster variant.

RAG Pipelines Under Adversarial Fire

A new Towards Data Science guide walks through building a small adversarial test set to surface retrieval failures before real users hit them. The approach targets the gap between standard evaluation metrics and the messy queries that break production RAG systems.

Speaker Recognition with Claude Code

Developers can follow a hands-on tutorial for building an internal speaker-recognition tool using Claude Code or Codex. It focuses on effectively driving agentic coding assistants to scaffold a working internal application.

AI for Decisions, Not Text

A separate piece profiles Jev, an AI system oriented around making decisions rather than generating prose. The framing pushes back on the text-generation default that dominates most current tooling.

PhD Workflows Get an AI Assist

Researchers get a practical breakdown of four uses for AI on a PhD thesis: finding citations, consolidating code, fact-checking, and preparing for the defense.

Hype Versus Reality

MIT Technology Review's newsletter covers pushback from Timnit Gebru and Emily Bender, who warn that recent breakthroughs and fears alike may be overstated. A companion essay argues that this summer's AI hype cycle deserves skepticism, noting Anthropic's late-April claims as one flashpoint.

Border Surveillance Failures

An MIT investigation documents deaths tied to the US "virtual wall" of surveillance towers, built at a cost of billions along the southern border.