NVIDIA's GTC 2024 (GPU Technology Conference) will be held from March 18–21.
This is perhaps the most highly anticipated event in the AI field currently: 1. It is arguably the most comprehensive and high-profile conference, featuring 1,045 sessions, including keynote speeches, panel discussions, and training events, with top figures from both industry and academia in attendance; 2. As the foundation for the entire AI infrastructure, every move NVIDIA makes is crucial; 3. We can sense future development trends through these activities.
While planning my schedule for online attendance (I had originally planned to attend in person but unfortunately missed out again due to a scheduling conflict), I used the filtering tools on the conference homepage to extract data for over 1,000 sessions categorized by content and industry. After translating them using Claude 3, I created two charts.
First Chart: A treemap showing the number of sessions across 16 major categories and 105 subcategories. For me, the next wave of AI applications is clear at a glance: Generative AI platform construction; AI inference; Metaverse and Digital Twins; Biomedicine; Robotics (focusing on synthetic data and simulation platforms)...
Second Chart: There are approximately 536 sessions related to specific industries (with some overlap), which is even more intuitive: Life Sciences; High-Performance Computing (HPC); Automotive; Manufacturing...
While processing this data, besides browsing through them manually, I didn't forget to test the capabilities of several models for the translation part—especially for the first table, which was quite long and not easy to handle accurately.
I tested Claude 3 Opus, GPT-4, and Gemini 1.5 Pro. Both Claude 3 and Gemini provided completely accurate responses, while GPT-4 was "lazy" again. Ultimately, because both Claude 3 and Gemini were accurate (correct counts, good translation), I simply chose the result from Claude 3.
Screenshot 1: Claude 3
Screenshots 2 & 3: GPT-4 and the code executed in the model's backend
Screenshot 4: Gemini 1.5 Pro
What can I say? OpenAI needs to step up.