State of AI, September 2025: A Broad Primer from an AI Team Lead in Education

Abstraction: Education-focused overview of AI history, capabilities, agents, and AGI trajectory

Key points:

• LLMs exhibit "jagged intelligence": can solve expert math yet fail simple comparisons; current state is text prediction not deep understanding
• Evolution: AI chatbots as oracles (2023) -> narrow agentic systems (2025) -> projected AGI (2027–2035 per most labs)
• Agents = LLM + system instructions + knowledge database + memory + tools; narrow-scope deployments most reliable
• Key limitations: no self-learning, context window as short-term memory, hallucinations harder to spot as models improve, context corruption at scale
• AI best applied in education for personalized tutoring at scale and accelerated customized learning programs
• AGI arrival would require rethinking economic systems; AI safety field aims to prevent misuse by bad actors

Connections: Chatgpt · Large Language Models · AI Agents · Artificial General Intelligence · AI Safety

Source: https://medium.com/@maksimkulichenko/state-of-ai-september-2025-a-broad-primer-from-an-ai-team-lead-in-education-12c27e5e9ce8