Microsoft’s New AI Models Could Pass PhD Exams: Insights from CTO Kevin Scott | Top Language Models | Large Language Models Tutorial Python Geeksforgeeks | Top 10 Most Popular Large Language Models | Turtles AI
Microsoft’s New AI Models Could Pass PhD Exams: Insights from CTO Kevin Scott
Microsoft CTO Kevin Scott has revealed that the company’s new artificial intelligence models could pass PhD qualifying exams, thanks to significant improvements in memory and reasoning abilities.
DukeRem
Microsoft predicts that new AI models will surpass OpenAI GPT-4’s reasoning capabilities and may even pass PhD qualifying exams.
In a recent event in Los Angeles, Microsoft CTO Kevin Scott discussed AI innovations and challenges with Reid Hoffman. Scott revealed that Microsoft’s new AI models might surpass PhD qualifying exams due to significant improvements in memory and reasoning abilities. The persistent memory, described by Scott as "episodic," allows AI models to follow the flow of communication with users, making them much more useful than previous versions. This development represents a step forward from the early days of ChatGPT and Microsoft Copilot, which were plagued by character limits and frequent hallucination episodes. Additionally, recent updates have made AI models more robust in reasoning, enhancing their ability to solve complex problems.
Scott illustrated that while GPT-4 performs like a high school student on AP exams, the new AI models could surpass PhD qualifying exams. Despite improvements in memory and reasoning, the real test lies in how we utilize this technology. Scott observed that, while in the past machine learning required graduate-level mathematical knowledge and thorough technical research, today, with available AI tools, even a high school student could accomplish complex tasks in one morning.
Scott’s remarks raise important questions about AI’s future impact. For instance, Elon Musk has predicted that AI could take over all jobs, providing goods and services efficiently and turning work into a hobby. This debate is part of a broader discussion on AI and its ethical and social implications.
Below are the technical details regarding the capabilities of the new AI models:
| Capability | GPT-4 | New AI Models |
| Persistent Memory | No | Yes |
| Reasoning Ability | High school level | PhD qualifying exams |
| Effectiveness in Financial Forecasts | High | Even higher |
| Robustness Against Hallucinations | Moderate | High |
Highlights
- Persistent Memory: New AI models can follow the flow of communication with users, significantly enhancing interaction.
- Reasoning Ability: Updated AI models can pass PhD qualifying exams.
- Use of AI Tools: Availability of AI tools allows even non-experts to perform complex tasks.
- Future Implications: The growth of AI capabilities raises ethical and social questions about work and productivity.
