🤖 AI Summary
The State of AI Report 2025, released by Nathan Benaich and Air Street Capital, is the annual, open-access roundup of breakthroughs, industry uptake, regulation, and safety trends—now in its eighth year and based on peer review plus a 1,200‑respondent practitioner survey. It documents a competitive frontier (OpenAI retaining a narrow lead while China’s DeepSeek, Qwen, and Kimi close the gap), rapid commercial adoption (44% of U.S. firms now pay for AI tools, average contracts ~$530K, and 95% of practitioners use AI), and the start of an “industrial era” marked by multi‑GW data centers and sovereign-backed compute projects where power supply is emerging as the main constraint.
Technically, 2025 was defined by advances in structured, verifiable reasoning: frontier labs combined reinforcement learning, rubric‑based rewards, and novel evaluation environments to produce models that can plan, reflect, self‑correct, and operate across longer time horizons. Embodied AI saw “Chain‑of‑Action” planning in robots (e.g., AI2’s Molmo‑Act, Google Gemini Robotics 1.5), and scientific AI moved from assistance to autonomous experimentation (DeepMind’s Co‑Scientist, Stanford’s Virtual Lab; Profluent’s ProGen3 suggests protein scaling laws). Politically, “America‑first” approaches, China’s open‑weights push, and stalled EU regulation reshaped geopolitics. Safety research shifted from abstract existential debate to practical reliability, transparency vs capability tradeoffs, and underfunded external oversight relative to frontier lab burn rates—raising governance and resilience as near‑term priorities.
Loading comments...
login to comment
loading comments...
no comments yet