Panel: Storage Systems in the LLM Era
summary
Panel argues that LLM training often tolerates “bigger and cheaper” storage but surfaces concrete problems: data provenance and tracking which data or model produced which output; rising multimodal volumes (James Webb ~235 GB/day), heavy pre-processing into tokens, and the boom of embeddings plus vector databases for retrieval-augmented generation. Practical topics include web‑crawl deduplication, read/write amplification from repeated preprocessing, telemetry needs, RLHF for tuning, conversational admin interfaces, and when AI actually helps autotune or transfer learn across storage systems.
AI-written summary · about
chapters (6)
- 0:00Panel introductions
- 6:27AI storage requirements
- 9:52Data management challenges
- 21:18AI for storage systems
- 28:14Natural language interfaces
- 38:09Education and skills
related talks
43mImportance of data in the time of LLMs
40mPannel Discussion | From Hype To Value: Real World Applications Of Generative AI And LLMs
36mTechniques to avoid common LLM pitfalls
40mLLMs & Data Warehouses: Reliable AI Agents for Business (Keynote)
30mLLMs: Beyond the Hype - A Practical Journey to Scale
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
28mThe Data That Shapes Foundational LLMs
22mPractical Strategies for Implementing Domain-Specific LLMs