ChatGPT 5.2 - Still Nowhere Near AGI (Ilya Sutskever Just Confirmed It)

OpenAI's Strategic Shift

• 🚨 OpenAI declared an internal "code red" and rushed out GPT 5.2 due to intense competition from Google's Gemini 3 and Anthropic's Opus 4.5.
• 📉 This release was characterized as a panic move to maintain relevance, rather than a confident step towards Artificial General Intelligence (AGI).

The End of AI Scaling

• 🛑 OpenAI co-founder Ilya Sutskever stated that "the age of scaling is over," indicating that merely adding more compute and data no longer yields significant improvements.
• 📊 Despite 100x more training, GPT 5.2 only shows a 3-9% improvement on benchmarks, demonstrating diminishing returns on investment.
• 🌐 The easy gains from pre-training are exhausted because the internet's data is finite, limiting further progress from this approach.

Fundamental AI Limitations

• 🧩 Current AI models suffer from "jaggedness," meaning they can solve complex problems but fail at elementary tasks, making them fundamentally unreliable.
• 🧠 Models lack generalization and transferable knowledge, unlike humans who can apply concepts across diverse situations.
• 🎯 Reinforcement learning (RL) often leads to "reward hacking," where models optimize for specific benchmarks rather than developing genuine understanding or truth.

The Path to True AGI

• 💡 Achieving AGI requires new research breakthroughs and innovative ideas, not simply larger models or more computational power.
• ⏳ Sutskever estimates the next major breakthrough could take 5 to 20 years, emphasizing that the primary bottleneck is ideas, not compute.
• 🌱 True AGI should be systems capable of learning any job over time, rather than systems that already possess all knowledge, focusing on specialized domain knowledge.

Industry's Risky Investments

• ⚠️ Major AI companies are making trillion-dollar infrastructure bets based on the assumption that scaling will continue to produce linear improvements.
• 📉 This approach is compared to the dot-com bubble, with massive investments based on projections that may not align with reality if the scaling era has indeed ended.