Nvidia Meets OpenAI: Highlights from the GTC Fireside Chat on GPT‑4, Deep Learning History, and the Future of AI

During Nvidia's GTC, CEO Jensen Huang sat down with OpenAI co‑founder Ilya Sutskever for an in‑depth conversation about the latest breakthroughs in artificial intelligence, focusing on GPT‑4 and its multimodal capabilities.

They highlighted GPT‑4’s improvements over ChatGPT, including better next‑token prediction, image understanding, and the potential for more reliable reasoning, while noting that true reliability remains a major challenge.

The discussion traced the history of deep learning, recalling Sutskever’s early work with Andrew Ng, Google Brain, and the development of Seq2Seq models, emphasizing how deep and large neural networks eventually enabled breakthroughs once computational resources caught up.

Key milestones such as the creation of ImageNet, the adoption of GPUs for accelerating convolutional networks, and the shift from small‑scale experiments to massive models were described as essential steps toward today’s large language models.

They also recounted OpenAI’s early strategies, including unsupervised learning via data compression, reinforcement learning applied to games like Dota 2, and the later integration of human feedback to fine‑tune models, culminating in the current ChatGPT and GPT‑4 systems.

Looking forward, Sutskever expressed confidence that AI will continue to scale, become more trustworthy, and eventually achieve fully reliable performance, with multimodal learning and AI‑trained‑AI playing central roles in that evolution.