Redefining Intelligence: How Reasoning Is Re-Shaping AI in 2025
summary
Claims that 2025 is the year of the agent and demonstrates OpenAI’s o1 family models that think before they speak, using test‑time computation instead of massive retraining. Shows a demo of the real‑time multimodal API: voice, screen sharing and image reading to diagnose a Japanese crop cutter, identify a worn friction pad and place a parts order. Reports jump from ~70% to 80% accuracy with runtime reasoning and higher jailbreak resistance (about 80% rejections versus ~20% for older models).
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:00Reasonings Goal
- 2:20Agenda
- 3:29Use Cases
- 4:44Models
- 6:55Test Time Compu
- 8:39Modalities
- 9:46Latency
- 13:00Agent Capabilities
- 14:23Agent Example
related talks
52mThe Agentic Shift- Are We Reasoning, Or Just Retrieving Smarter?
25mFrontier AI Models: A Review of Past Nine Months
31mKeynote: Noam Brown, Research Scientist at OpenAI in Conversation with Joe Spisak
01h 17mThe year of the AI Agent: Beyond Automation
30mHow AI Agents Will Change Everything (2026)
31mMachine Reasoning and System 2 Thinking
01h 01mKeynote Panel: Future of AI agents
17mRethinking AI Safety in the Age of Agentic Systems