ChatGPT o3 reshapes what AI can actually do

In a significant leap forward for artificial intelligence, OpenAI has unveiled ChatGPT o3, showcasing capabilities that push well beyond what we've come to expect from AI systems. The new model doesn't just think with text—it thinks with images, remembers conversations over time, and demonstrates problem-solving abilities that approach genuine research capabilities.

Key developments in o3's capabilities

• Multimodal reasoning with images: The model can now analyze images with remarkable precision—reading barely visible signs, identifying film locations, finding Waldo in a crowded scene, and even determining guitar chords from photos.

• Enhanced memory and personalization: O3 retains information from previous conversations and uses it to provide personalized recommendations and learning pathways, effectively becoming a thoughtful tutor tailored to your knowledge gaps.

• Significant cognitive improvements: The model has jumped from scoring 8% to nearly 25% on "Humanity's Last Exam"—a collection of questions specifically designed to be unsolvable by AI at the time of creation—representing a threefold improvement in just months.

Breaking the knowledge barrier

Perhaps the most significant development is o3's apparent ability to push beyond what I call the "knowledge ceiling" of AI. Previous models were limited to information contained within their training data—essentially regurgitating human knowledge in new combinations. O3 demonstrates an ability to work through incomplete research papers and derive conclusions that weren't explicitly provided, suggesting it might be capable of genuine discovery.

This shift matters immensely as it potentially transforms AI from a tool that enhances human productivity to one that independently advances human knowledge. If these capabilities continue developing at the current pace, we could see AI making autonomous contributions to drug design, sustainable agriculture, clean energy, and longevity research within years rather than decades.

What OpenAI isn't telling us

While the capabilities shown are impressive, there's significant opacity around o3's development and benchmarking. Unlike academic research, which undergoes rigorous peer review, many of OpenAI's claims about o3's intelligence remain unverified by independent researchers. Claims about "genius-level IQ" should be approached with healthy skepticism until independently verified.

Additionally, the demonstrations shown likely represent optimal outcomes rather than typical performance