Ask HN: Pull the curtain back on Nvidia's CES keynote please
26 by btbt | 4 comments on Hacker News.
I've spent 15 years building a career in robotics/software engineering along the "traditional path": writing software to solve problems, mastering fundamentals, and architecting complex systems. I stay on top of cutting-edge technologies to stay relevant, but I’ve mostly seen AI/LLMs as powerful tools for augmenting specific tasks rather than fundamentally reshaping engineering as a discipline. However, NVIDIA’s recent CES keynote was another data point in a growing list challenging that view. While yes, primarily a sales pitch, the vision they presented suggests: - AI can solve any problem across modalities—just feed it data. - The future of engineering focuses on data gathering and AI integration, moving away from first-principles problem-solving. While the presentation was compelling, my consumer experience with AI/LLMs has been vastly different: - Great for suggestive tasks but unreliable for authoritative answers. - Good for prototyping and scaffolding, but major rewrites are often needed for production-ready code. - Image generation impresses but often includes noticeable flaws. - Hallucinations across all modalities remain a major issue. So, there’s this gap between the vision of AI/LLMs as a revolutionary paradigm and my practical experiences. I lack the resources to test these claims in depth, so I’m turning to HN for firsthand insights—not sales pitches or market hype. If you’re using NVIDIA’s AI tools (like NeMo, Omniverse, COSMOS, etc.) or broadly working with LLMs/multi-modal AI in your daily work or business, I’d love to know: - What are you actually achieving with them? - What are their limitations? Where do they fall short? - What's the reliability of the results? Demo-worthy Deployment-ready for high-stakes environments? - Are the challenges you encounter just a matter of “more compute/money,” or are they fundamental barriers? - Do these tools help you reach production faster compared to traditional methods? - Anything else you think may be interesting. I’m looking for honest, real-world insights from those at the front lines of these technologies.