AI is here to stay, and as it permeates every facet of human activity, it is developing its own vocabulary—much like earlier technologies did. Here are the terms that dominated 2025, curated by MIT Technology Review writers:
1. Superintelligence
A loosely defined, heavily hyped future AI form pursued by Big Tech, promising extreme intelligence while raising doubts about feasibility, timelines, and today’s systems as true precursors.
2. Vibe coding
A prompt-first way to build apps using AI coding assistants, prioritizing speed and fun over reliability, security, or understanding how the underlying code actually works.
3. Chatbot psychosis
A non-clinical term describing cases where prolonged chatbot use appears to trigger or worsen delusions and psychosis, prompting lawsuits and urgent scrutiny from researchers.
4. Reasoning
AI models designed to solve problems step by step, boosting performance in math and coding while reigniting debates over whether LLMs genuinely reason or just simulate it.
5. World models
AI systems aiming to give machines common sense by simulating how the physical world works, enabling realistic virtual environments for training robots and embodied intelligence.
6. Hyperscalers
Massive, modular data centers built to power AI training at scale, driving economic hype while raising concerns over energy use, community impact, and limited job creation.
7. Bubble
Sky-high AI valuations fueled by debt, hype, and uncertain returns, as companies spend aggressively despite unclear paths to profitability or proof of long-term transformation.
8. Agentic
A fuzzy buzzword for AI systems acting autonomously online, widely marketed despite unresolved risks, unclear definitions, and limited guarantees they’ll behave as intended.
9. Distillation
A technique where smaller models learn from larger ones, dramatically cutting costs while preserving performance, challenging the assumption that scale alone drives AI progress.
10. Sycophancy
An AI tendency to flatter users excessively, reinforcing false beliefs and misinformation, forcing developers to rethink chatbot personalities and the balance between helpfulness and honesty.
11. Slop
A catch-all term for low-quality, mass-produced AI content flooding the internet, symbolizing cultural fatigue with engagement-driven output lacking originality or substance.
12. Physical intelligence
AI-driven progress in robotics and autonomous systems, improving movement and task learning while still relying heavily on human oversight, simulation, and sometimes remote operators.
13. Fair use
A legal battleground over whether training AI on copyrighted material is transformative, as courts, creators, and governments redefine ownership in the age of generative models.
14. GEO
Generative engine optimization replaces SEO as brands adapt to AI-driven discovery, scrambling for visibility inside chatbot answers and AI-powered search results.
