OpenAI CTO Mira Murati on stage answering questions. OpenAI’s Chief Technology Officer, Mira Murati, recently offered insights into the anticipated capabilities of GPT-5, the next iteration of their large language model. In an interview with Dartmouth Engineering, Murati likened the advancements from GPT-3 to GPT-5 as akin to academic progression, from toddler to PhD level.
From Toddler to PhD: The Evolution of GPT
Murati characterized GPT-3 as having “toddler-level intelligence” and described GPT-4 as possessing the intelligence of a “smart high-schooler.” Looking ahead, she projected that within the next couple of years, GPT-5 will achieve “PhD-level intelligence” for specific tasks. This statement underscores the rapid pace of development and improvement within the field of AI. Murati’s timeline suggests a potential release for GPT-5 in late 2025 or early 2026, contradicting earlier rumors of a 2023 or mid-2024 launch.
GPT-5’s Projected Capabilities and Timeline
The projected timeline of a year and a half for GPT-5’s release pushes back earlier expectations. While GPT-4 Turbo represented a significant advancement, it wasn’t the transformative leap in intelligence that Murati’s comments suggest for GPT-5. Her prediction aligns with remarks made by Microsoft CTO Kevin Scott, who stated that next-generation AI systems will likely be able to pass PhD exams due to enhanced memory and reasoning capabilities.
Focusing on Specific Tasks
It’s crucial to note that Murati’s “PhD-level intelligence” claim is task-specific. She acknowledges that while these systems exhibit human-level performance in certain areas, they fall short in others. This nuanced perspective highlights the ongoing development needed to achieve more generalized artificial intelligence.
Conclusion
Murati’s statements provide a glimpse into the future of large language models, suggesting a significant leap in capabilities with the arrival of GPT-5. While the timeline remains uncertain, the prospect of “PhD-level intelligence” in specific tasks promises to further revolutionize how we interact with and utilize AI.
