Microsoft has come out with its report card for year 2023. In a blogpost, it says, in 2023, researchers and technology enthusiasts witnessed a groundbreaking moment when they engaged with GPT-4, a large language model (LLM) developed by OpenAI. AI experts, including Ece Kamar from Microsoft, were astonished by the unprecedented capabilities displayed by GPT-4, prompting discussions about the potential emergence of artificial general intelligence. Throughout the year, AI advancements took center stage in public discourse, captivating both technology leaders and the general public. The question of whether we are witnessing sparks of artificial general intelligence lingered, but one thing became certain—we entered the era of AI, bringing profound changes to work and daily life.
AI’s Everyday Innovations:
AI transitioned from the research domain to everyday applications accessible to millions. Services like ChatGPT and AI Copilots, designed to assist with complex tasks, found their way into the mainstream, weaving into business software and services.
Strengthening AI Foundations:
Microsoft played a pivotal role in redefining the state of the art across AI areas, focusing on model efficiency, performance, and capability. Researchers introduced Retentive Networks (RetNet), an alternative to the transformer architecture, and the 1-bit transformer architecture called BitNet for more computationally efficient language models.
The Phi family of small language models expanded with Phi-2, a 2.7 billion-parameter model, showcasing advancements in reasoning and language understanding. Models like Orca and Orca 2 demonstrated how improved training methods can elevate small models’ reasoning to match larger counterparts.
Advancing Model Usage:
Frameworks like AutoGen simplified the orchestration, optimization, and automation of LLM workflows, enhancing the creation of LLM-based applications. Medprompt showcased the efficiency of general foundation models over specialized models through advanced prompting strategies.
Best Practices for Responsible AI:
Microsoft prioritized responsible AI development, supporting the White House’s voluntary AI commitments. Researchers examined model capabilities and limitations, addressing issues related to AI harms, and emphasized transparency as a cornerstone of responsible, human-centered AI.
Scientific Exploration with AI:
AI played a crucial role in accelerating scientific discovery. From enhancing subseasonal forecasts using machine learning to predicting protein structures with greater accuracy through Distributional Graphormer (DIG), Microsoft’s AI advancements spanned diverse disciplines.
MatterGen, DeepSpeed4Science, and other initiatives demonstrated the potential of deep learning in revolutionizing natural sciences, offering tools to unlock mysteries in fields like drug development and renewable energy.
Empowering Individuals and Society:
Microsoft’s commitment to positive human impact was evident in AI applications in health and education. Researchers explored the impact of LLMs in healthcare, as seen in the book “The AI Revolution in Medicine: GPT-4 and Beyond.”
AI in radiology, Shiksha Copilot assisting teachers in India, and efforts to improve models on non-English languages through Project VeLLM showcased Microsoft’s dedication to equitable and accessible AI technology.
In summary, 2023 marked a pivotal year in AI, where GPT-4’s capabilities sparked discussions about the future of artificial general intelligence. Microsoft’s contributions across model development, responsible AI practices, scientific exploration, and societal empowerment positioned it at the forefront of AI’s transformative journey. As we reflect on the progress of the past year, the stage is set for even greater advancements in 2024.
