AI & ML Research 24 Hours

Last updated: September 23, 2026, 6:06 AM ET

OpenAI Ships GPT-6 Sol and Luna

OpenAI launched GPT-6 Sol and Luna, a pair of models the company positions as bringing frontier intelligence to everyday use. The dual release splits capability across two tiers rather than a single flagship, a pattern that has become standard among major labs.

RAG Pipelines Under Adversarial Pressure

A new Towards Data Science piece argues teams should break their own RAG pipeline before users do, offering a small adversarial test set that surfaces the retrieval failures standard evaluation suites miss. The approach targets the gap between benchmark scores and the messy queries that reach production.

Building Internal Tools with Claude Code

For engineers wiring up speech systems, a walkthrough shows how to build a speaker-recognition app using Claude Code or Codex. The tutorial focuses on shipping an internal tool quickly rather than optimizing a research-grade model.

AI in Academic Workflows

A separate guide outlines four ways to use AI on a PhD thesis: finding citations, consolidating code, fact-checking, and preparing for the defense. The author frames these as leverage points where models save real hours without replacing the researcher's judgment.

Decision AI Beyond Text Generation

An introduction to Jev examines an AI built for making decisions rather than generating text, a departure from the chat-centric framing that dominates current tooling.

Skepticism on the Summer of Hype

MIT Technology Review's Download newsletter questions whether the latest breakthroughs and fears are more hype than reality, citing Gebru and Bender's warnings about overstated claims. A companion piece urges readers not to be fooled by the summer of AI hype, noting Anthropic's late-April claims as a case in point.

Surveillance Tech's Deadly Record

MIT Technology Review's investigation into the virtual border wall reports that billions spent on surveillance towers along the US border have produced deadly failures, a cautionary case for engineers deploying automated monitoring at scale.