The Information's TITV GPT-5.2 vs. Gemini 3 Pro, Larry Ellison’s Big Bet & Risk, Nebius CBO on AI Cloud War | Dec 12, 2025
45 snips
Dec 12, 2025 Join Roman Chernin, Co-founder of Nebius, an AI-focused cloud provider, as he dives into the evolving landscape of AI infrastructure. He reveals how open-source models can drive performance while discussing the trade-offs between closed and open-source systems. Roman also highlights Nebius' dynamic pricing strategies and the emphasis on total cost of ownership for customers. With ambitious goals for 2026, he outlines how specialization can outshine universal cloud solutions in the competitive AI sector.
AI Snips
Chapters
Transcript
Episode notes
Models Showing Capability Awareness
- GPT-5.2 shows stronger capability awareness, asking whether it has enough info and when to bring humans in.
- Niko Grupen found this emergent self-scrutiny especially valuable for legal workflows.
Frontier Models Are Converging
- Frontier models (GPT-5.1/5.2, Sonnet 4.5, Gemini 3 Pro) are converging on similar mean performance.
- Niko Grupen emphasized that task-specific sharp edges matter more than small mean-score deltas.
Evaluate Models With Real-World Benchmarks
- Build benchmarks that reflect real-world usage to judge model progress accurately.
- Use emergent behavior and customer feedback, not only mean scores, to decide releases.