GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in Open Source
Download as PPTX, PDFā¢0 likesā¢227 views
The document discusses the democratization of Artificial General Intelligence (AGI) through open-source community efforts, emphasizing its significance like that of music and mathematics. H2O.ai is highlighted for its role in various sectors, using data science and machine learning to drive innovation and improve public services, while also showcasing the achievements of its expert community. The overarching theme is the need for responsible AI practices and the integration of AI applications across multiple industries to foster collaboration and drive societal benefits.
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in Open Source
- 2. AGI - The making of the worldās most powerful AI will be Open Source and Community led
- 3. Like Music, Math is generational wealth of the human species!
- 4. Like Math, AGI will belong to the Community in Open Source!
- 5. Open Source is freedom, not free!
- 6. Open Source is the defense of the community with code
- 7. Open Source is the defense of the community with AI
- 8. Data Scientist = 100% data Business Unit = 85% data, 15% storytelling Executive = 15% data, 85% storytelling Board = 99% storytelling GrandPa = 100% storytelling Our (LLMs) journey to here
- 9. Time is the only non-renewable resource
- 10. H2O.ai Confidential Democratize AI with H2O.ai 8 7 222OF THE H2O OF THE TOP 10 BANKS OF THE TOP 10 4 OF THE TOP 10 MANUFACTURING COMPANIES INSURANCE COMPANIES FORTUNE 500 CUSTOMER OBSESSION MAKER CULTURE Commonwealth Bank, Goldman Sachs, Wells Fargo, NVIDIA, Capital One, Nexus Ventures, New York Life 2M Community 30 Kaggle Grandmasters 2012 FOUNDED IN SILICON VALLEY $256M FUNDRAISED, SUSTAINABLE
- 11. H2O.ai Confidential H2O.ai Confidential H2O products are backed by 10% of the Worldās Data Science Grandmasters and a Team of Experts who are relentless in solving critical problems. KAGGLE GRANDMASTERS Worldās Best Data Scientists 2 Quadruple GMs 1 Triple GM 7 Double GMs #1 KGM #3 KGM 5Top 10 Globally 10 Kaggle Masters H2O.ai KGM HIGHLIGHTS ā2023 H2O KGMs win 3 healthcare-related competitions āJuly 2023 H2O KGMs make āTop GenAI Scientistsā list āOct 2023 H2O paper accepted at EMNLP 2023 āNov 2023 H2O KGMs win 1st place in Kaggle Science Exam competition āJan 2024 H2O KGM places 2nd in Detect AI Text competition āFeb 2024 introduced foundational model H2O- Danube 1.8b
- 12. BC is Before Covid
- 13. Every Nation needs to be an AI Nation
- 14. Prompt is your intellectual property
- 15. Everyone needs their own GPT
- 16. Every organization needs to own its GPT
- 17. Open source crushed the GPT trademark
- 18. Open source crushed GPT trademark, not fully! https://tsdr.uspto.gov/documentviewer?caseId=sn97733259&docId=FREF20240206125856&linkId=1#docI ndex=1&page=1
- 19. Democratize AI
- 20. Democratize AGI
- 21. Gaia Benchmark for General Intelligence agents OS-Copilot / FRIDAY Autogen Coding General Assistants SWE-Bench OpenDevin Eureka https://gaia-benchmark-leaderboard.hf.space/?__theme=light
- 22. dream
- 23. AI for Good
- 24. AI for Good AI for Climate Change AI to Protect WildLife and Pets AI To Manage Pandemic Supply Chain AI to Model WildFire Behaviors AI to Predict Hurricanes AI for H2O (Water) AI to Democratize Health AI to Educate and Upskill our Public Services Responsible AI and Fairness
- 26. AI to predict Hurricanes
- 27. AI to fight Climate Change
- 28. AI to democratize health
- 30. Co-create
- 31. H2O.ai Confidential Generative AI Predictive AI Generative AI AWARD-WINNING AUTOML reaches a new audience with chat & function calling assistants for ML Enhanced AI for Documents Automation for data labeling Automation of code for AI Apps CUSTOMGPTS: H2O foundational models, fine-tuned models, personality, on top of predictive models Convergence of Worldās best Predictive and Generative AI GenAI Apps Enterprise h2oGPTe: Python APIs and UI for chat, RAG customGPTs Function-calling & Agents Prompt Tuning Foundational Models document and data labeling on-premise, air-gapped, cloud VPC | SaaS +
- 32. H2O.ai Confidential GenAI AppStudio Structured Datasets Unstructured Datasets ETL / Prep for LLMs Documents ā QA Pairs Fine Tuning LLMs (& Prompts) End Users Vector DB (Embedding s) myGPT R. A. G. Talk to your Data Document QA Document Chat Image/Video Chat LLM Query GenAI Apps + + + + + + LLM Data Studio AI Engines MLOps EvalStudio AI Apps + LLMs Integration LLMOps API Prompt Studio Continuous Feedback Parsing . Chunking Indexing . Embeddings LLM Agents Enterprise h2oGPTe (RAG) Make Your Own Eval Predictive AI Layer Predictive + Gen AI h2o-functions Worldās Best AI to do Gen AI Tabular Data
- 33. Raise a forest, not just a tree. Make ecosystems
- 34. Tech Ecosystem of H2O GenAI OPEN SOURCE MODELS VECTOR DATABASES CLOSE SOURCE MODELS LLM INFERENCING HARDWARE MODEL HOSTING CLOUD PROVIDERS
- 35. Algorithms Design Data Domain Business Applications Data is a Team Sport Teamwork makes the Dream Work
- 36. Thank you, Community. Gratitude. Makers, past, present and future! H2O Movement Community and Customers! To all greatness, our greetings endaro mahanubhavulu
- 37. LLM-Based Evaluation Prompt/Context and Document Embedding Dimensionality Reduction and Clustering Cluster/Topic Summarization Auto Prompt Generation Prompt Perturbation ā¢ Invariance ā¢ Adversarial Robustness ā¢ Performance Drop Stratified Sampling Weakness Detection ā¢ Weak clusters/topics Resilience ā¢ Performance under distribution shift Interpretability Outcome Analysis: Cluster Eval Input-Output Visualization Diagnostics
- 39. Open Source Foundation Models H2O Danube 2 LLM 1.8B, 2T Tokens, 4K seq length fp8 precision during training (quantization) Grouped Query Attention (Dropped Windowing; Mistral Tokenizer) Compute 7,600 GPU hrs, (8xH100, 45 days) Inference is efficient. Embeddable. Cost efficient! (lot less than you think!) Ready to pre-train on your data! Democratizing Foundation LLMs
- 41. Using LLM Studio DPO, with good chat data Fine-Tuned for Chat
- 42. Guardrail LLMs and Gateway LLMs LLMs to safeguard an LLM
- 43. Costs and Latencies of AI Training Costs Inference Cost & latency vLLM Data Labeling, Synthetic Feedback Private On-prem, behind your firewall Sovereign Data Sovereign AI
- 44. PII Detection Detect patterns of personal identification https://www.kaggle.com/competitions/pii-detection-removal-from-educational- data/discussion/481135 LLM Generated Content Detection Easier to detect human generated Prompt Recovery and Intent discovery Your own post-trained, fine-tuned content Post-train and fine-tune LLMs on small hardware Early SLM (Danube 2) applications
- 45. Open Source h2oGPT Fine-tuning and RAG of Best-in-class Open Source LLMs Open Source H2O LLM Studio Best-in-class Studio to fine-tune LLMs
- 46. Enterprise h2oGPTe Best-in class RAG and API to power GenAI Apps
- 47. H2O EvalStudio Evaluation-led building of LLM-powered Applications Open Source H2O EvalGPT Evaluate LLMs for public and private benchmarks
- 48. Converging Predictive AI and Generative AI
- 49. H2O Agents Converging Predictive AI and Generative AI
- 50. H2O Document AI Prompt powered AI for Documents
- 51. H2O Label Genie ETL for GenAI
- 52. H2O Document Diff Document Difference Engine powered by GenAI
- 53. Data sanskrit. plural of datum. to give. to impact with energy. mineral rich in energy.
- 54. AI will usher in abundance
- 55. AI will usher in abundance abundance of time abundance of space outer space, inner space abundance of matter & energy
- 56. With great powers, comes greatā¦ Responsible AI.
- 57. BC is Before Covid
- 58. Long time ago, there once were - Virus, Wars and Superstitions
- 59. Long time ago, there once were - Virus, Wars and Fake News
- 60. AI can make a difference.
- 61. We can make a difference.
- 62. busy being born and busy dying. - Bob Dylan
- 63. this will be fun!
- 64. Gratitude
- 66. Make data and AI your first-class assets
- 67. Thank You: GenAI Open Source Ecosystem h2oGPT (Open Source GPTs need RAG) PyMuPDF BGE, Instructor embeddings ragas, evalGPT.ai, hf/leaderboard, llm-eval hnswlib, Milvus, Chroma LangChain, LlamaIndex, Open Interpreter vLLMs, tgi-0.9 HuggingFace, H2O LLMStudio, GGUF, AWQ, Gradio Alpaca LoRA, llama.cpp, AutoGPTQ, hf/TheBloke OpenAssistant LLama2, Falcon, Mistral, GPT3, Eleuther, BigScience Pytorch, DeepSpeed āAttention is All You Needā team and Transformers CUDA, RocM Python, Go, Rust, K8s, Clouds # others I might have missed # @ylecun # @ Dr. Ebtesam Almazrouei # @tloen/alpaca-lora @ecjwg
- 68. ETL for LLMs H2O Data Studio Label Genie powered by LLMs Hydrogen Torch Labeling as a Service
- 69. Evaluation led LLM Development evalGPT.ai ragas, retrieval Evaluation as a Service Eval Studio
- 70. When Generation is abundant Curation becomes valuable
- 71. AI to curate AI
- 72. AI to do AI
- 73. customer is the moat.
- 74. Co-creation
- 75. Gov GenAI App Store
- 76. RAG + Gen AI Use cases Customer Experience, Contact Centers Document Processing Generating Marketing Generating Code and Apps Procurement, RFPs, Contracts, Inventory Portfolio Recommendation (Earnings Calls) Custom GPTs Meeting Summaries, Tumor Board Summary Multi-modal, Speech Document Extraction, ETL for LLMs Labeling, Translation Ask Data. BI.
- 77. H2O.ai Confidential AI for ALL AI for Sales People AI for Investors AI for Manufacturers AI for Retailers AI for Hospitals AI for Transportation AI for Urban Planning AI for Financial Services AI for Real Estates AI for Energy AI for Tourism AI for Hospitality AI for Oil and Gas AI for Nature AI for Future AI for Good Government Scam Shield LLM based Scam Prevention Service LLM Governments Virtual Advisor LLM based conversation support services LLM Governments Blueprint Designer LLM based design and construction LLM Tourism Tour Personalization LLM based travel and tour recommendations LLM Tourism Language Assist LLM based multilingual assistance LLM Governments Waste Management GenAI based waste optimization LLM Smart City Dynamic Road Tax Services to provide route recommendations LLM Manufacturing Green Builder LLMs for Carbon reduction plans LLM Hospitals Healthcare AI Best public health advice using GenAI LLM Hospitals Disease Advice LLMs for patient care and disease advice LLM Real Estate Tenancy Screener LLM for person risk profiling LLM Security Edge Surveillance Local LLMs in drones for Intelligent Surveillance LLM Gen AI App Store - Applications Powered by LLMs
- 78. H2O.ai Confidential AI App Store - Everything AI in one place AI for ALL AI for Sales People AI for Investors AI for Manufacturers AI for Retailers AI for Hospitals AI for Transportation AI for Urban Planning AI for Financial Services AI for Real Estates AI for Energy AI for Tourism AI for Hospitality AI for Oil and Gas AI for Nature AI for Future AI for Good
- 79. + Scaling the AI & GenAI for 100s of Use Cases Data LLMs / Models AI + LLM Apps GenAI App Store Model Deployments ATL AI + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + AI Access Regions 100s of Use Cases
- 80. AIaaS Ecosystem with H2O.AI at AT&T AI Governance Center Any AI/ML library Build your own models Import Models Create App/ Deploy Model Microservice ā Model Validation ā Adversarial analysis ā Backtesting ā What-if analysis ā Suggest models ā Model catalog ā Suggest features ā Feature store ā Ensemble strategies ā Model benchmarking Create AI App Store Model Store H2O Feature Store App Store Model Analytics Platform AI Bias eval, MLI AI Ops Application Deploy AI 1 2 3 Self Service Data Ingestion H2O - 3 Data Store on Prem H2O Olympics 2.5 H2O Olympics Responsible AI AI/ML User Model Store H2O WAVE H2O WAVE H2O Feature Store HDFS Cloudera H2O MLOps Model Repository Model Deployment Model Monitoring High Availability Model Hosting GenAI (LLMs) Integration 4
- 81. Customer and Community Love is the greatest force in nature.
- 82. From Data Science to Strategy Science
- 83. Silence The quieter you become the more we can hear Ram Dass
- 84. listen silent
- 87. you can do anything you set your mind to. eminem.
- 88. Courage
- 89. whose world is this?
- 90. https://www.youtube.com/watch?v=ahY62u62Uw8 Transform manager to a leader caterpillar to a butterfly!
- 91. Rise of Data Scientist From Data to Strategy Science CDAO now works closely with CEO
- 92. Intelligence is the Edge
- 93. The only wisdom we can hope to acquire Is the wisdom of humility: humility is endless TS Eliot said that.
- 94. You can be in my dream If I can be in yours. Bob Dylan said that.
- 95. You can be in my selfie If I can be in yours.