Keynote: Open Source in the Age of AI
summary
Open source is driving AI efficiency: open models like DeepSeek close the quality gap to closed systems while cutting training costs, frontier runs hit ~billion‑dollar scale (GPT5 cited), and hyperscalers plan >$300B capex—exceeding Portugal’s GDP—while inference demand rises ~50×. Software like VLLM and DeepSpeed claim 4×–6× or more GPU/inference efficiency, enabling many smaller fine‑tuned models instead of one giant model and making agentic, multi‑agent workloads economically feasible.
AI-written summary · about
chapters (17)
- 0:00Welcome
- 1:04Conference logistics
- 2:31Open source and AI trends
- 4:37AI investment in 2025
- 7:30AI bubble concerns
- 9:17AI beyond a tool
- 10:23AI energy demands
- 13:40AI infrastructure race
- 14:48Open source in AI
- 16:36Open source models lower costs
- 18:03Efficiency in AI models
- 19:29Enterprise open source models
- 20:55Linux Foundation AI projects
- 22:43Next stage of AI
- 23:25Agentic AI rise
- 24:30Open source and Agentic AI
- 25:37Sustaining AI growth
related talks
13mKeynote: Introductory Remarks
16mKeynote: Open Source for AI and AI for Open Source
27mKeynote: State of the Union
54mThe Future of AI: Building the Most Impactful Technology Together
11mKeynote: Open Source Stack Empowers the AI Era
16mKeynote: Open Source AI Everywhere
13mKeynote: State of Open GenAI
33mKeynote: Fine-tuning our way towards openness in AI