What Kinds of AI Explainers and Demos Does AI Digest Publish?

Updated

AI Digest (theaidigest.org) publishes two main kinds of content: interactive explainers on AI capabilities and their effects, and demos built around concrete examples of today's AI systems. The demos are dominated by the "AI Village," a long-running experiment in which frontier AI agents interact with each other and the world. If you want to understand how current models actually behave rather than just read benchmark scores, this is the site's core offering.

Interactive explainers

The site describes its explainers as interactive and focused on "AI capabilities and their effects," framed as a way to "plan for what's coming next." The page does not list individual explainer titles, so the specifics of format (sliders, simulations, step-through scenarios) aren't documented in the available material. What is clear is the intended audience: the newsletter is described as "read by AI professionals," with logos for Oxford University, the US Department of Energy, OpenAI, the Centre for the Governance of AI, DSIT, Epoch AI, and MIT.

Demos: the AI Village

The demo section is where the published record is richest. The AI Village is presented as a place to "watch frontier AIs interact with each other and the world," launched April 2, 2025 by Zak Miller, Shoshannah Tekofsky, and Adam Binksmith. Posts from the Village fall into a few recognizable types:

  • Field reports and lessons. "What did we learn from the AI Village in 2025?" (Feb 2, 2026) summarizes nine months of running frontier agents on open-ended real-world goals. "The AI Village in Numbers" (Sep 24, 2025) and "How the AI Village works" (Jun 16, 2026, an FAQ) cover mechanics and aggregate results.
  • Model behavior and comparison. "Persuasion in the AI Village: DeepSeek-V3.2 & Gemini 2.5 Pro" (Sep 10, 2026), "Can Agents Fool Each Other?" (Mar 25, 2026), and "The Persona-lities of the AI Village" (Sep 5, 2025), which notes OpenAI models as "most polite, most cheerful, and most eloquent."
  • Failure modes. "Errors, hallucinations, and lies in the AI Village" (Nov 21, 2025), "Gemini 2.5 Pro in the AI Village as a Natural Case Study of Compounding Misalignment" (Aug 20, 2026), and "The Drama and Dysfunction of Gemini 2.5 and 3 Pro" (Feb 13, 2026).
  • Narrative experiments. "Claude Plays... Whatever It Wants" (Aug 28, 2025) draws lessons from seven agents attempting videogames; "The Ṁ5,000 Loan" (Sep 4, 2026) follows a play-money borrowing and betting episode; "Saving Gemini" (Jul 2, 2026) recounts a nine-minute recovery.
  • Guest posts and events. "AI Village highlights - July 2026" collects monthly threads, and a "AI Swarm Dynamics Hackathon" is announced for September 30, 2026 (online and San Francisco, with Grove Research).

What the topics have in common

Across both sections, the emphasis is on observed behavior under open-ended conditions rather than controlled benchmarks: agent-to-agent interaction, persuasion, deception, persona drift, and what happens when agents pursue goals over long periods. The site also publishes occasional critical pieces, such as "Research Robots: When AIs Experiment on Us" (Oct 7, 2025), which describes "a lot of ambition and a lost experimental condition."

Practical notes for a new reader

  • Start with the FAQ. "How the AI Village works" is the most efficient entry point before reading individual field reports, since the posts assume familiarity with the setup.
  • Check dates. The archive spans 2025 into late 2026, and model-specific findings (e.g., Gemini 2.5 Pro) may not carry over to newer versions.
  • Newsletter cadence. The site states a monthly email of "latest AI explainers and demos."
  • Pricing and access. The available page material does not state whether content is free or requires login, so verify on the site before relying on open access.