The Parrot in the Machine (www.nybooks.com)

What happened: The piece traces modern AI back to Claude Shannon’s 1950 letter‑guessing experiment and shows how today’s systems are essentially scaled-up versions of that idea: massive, probabilistic next‑token predictors trained on terabytes of text. The release of ChatGPT in November 2022 turbocharged public uptake (100M users in two months), spawned enormous VC and corporate investment (Microsoft’s multi‑billion backing, global AI funding >$100B), and prompted projects like OpenAI’s Stargate data‑center build and a proposed always‑on wearable companion with cameras and mics. The market surge has made AI ubiquitous in drafting copy, code, and student essays while also powering energy‑hungry infrastructure. Why it matters and technical implications: For the AI/ML community this is a reminder that high fluency doesn’t equal understanding. Chatbots produce human‑sounding text by extending statistical patterns, which explains their strengths (boilerplate, code synthesis) and failure modes (hallucinations, invented citations, plagiarism, flattened creative texture). The story highlights systemic issues: data provenance, model evaluation, grounding to factual sources, compute and energy costs, and socio‑technical impacts on education, jobs, and misinformation. Voices from researchers (Bender & Hanna’s “synthetic text extruding machines”) to CEOs (claims of imminent superintelligence) underscore the mix of hype and real capability—calling for technical work on grounding, citation, alignment, and policy responses as these probabilistic systems scale.