Industry Panel
summary
Panelists argue concrete shifts: PyTorch democratized AI by making training ubiquitous while model deployment is a dark art that still blocks smaller teams; inference demand scales with public use so GPU fleets and latency matter as much as training. Expect multi-vendor chips, tighter hardware–software co‑design, and a talent bottleneck around compilers, GPU kernels, and systems. Data sourcing and cleaning rise as the next hard constraint, and open releases like Llama 3 cost a lot to build but accelerate the ecosystem.
AI-written summary · about
chapters (11)
- 0:00Welcome and panelists
- 1:51Personal journeys into AI
- 3:35AGI and intelligence thresholds
- 6:06Scaling laws and AI trends
- 7:51Breaking into AI
- 11:23AI replacing mundane tasks
- 13:49AI infrastructure challenges
- 16:15Data bottlenecks and training
- 17:41Regulation and energy in AI
- 18:23PyTorch and open source
- 23:20Closing remarks
related talks
33mLive Q&A Session 2
01h 00mPANEL - Data, Code, and Life with LLMs.
32mPanel: The State Of AI: Sustainable Or A Bubble Waiting To Burst?
31mPanel: The Future of AI
47mBeyond Agents: What AI Strategy Really Needs in 2025
58mPanel Discussion: AI- Friend or Foe?
48mCollaboration Synthesis:Engineering India’s Open AI Future from Classroom to Cloud
05mFrom One Node to Distributed Training and Inference. How the PyTo...