* 官方 Twitter: https://x.com/aiDotEngineer * 議程: https://www.ai.engineer/worldsfair/2024/schedule * 直播錄影 * Keynotes & CodeGen Track: https://www.youtube.com/watch?v=5zE2sMka620 * GPUs & Inference https://www.youtube.com/watch?v=JVSKlEmUr0k * Keynotes & Multimodality Track https://www.youtube.com/watch?v=vaIiNZoXymg * Open Models https://www.youtube.com/watch?v=R0X7mPagRiE > ✅ 表示我有製作影片截圖和逐字稿,可以快速閱讀 ## Day 1 Keynote * Open Challenges for AI Engineering (simon) * ✅ https://ihower.tw/watch/aief-2024-keynote-day1-open-challenges-for-ai-engineering/ * 作者自己的講稿: https://simonwillison.net/2024/Jun/27/ai-worlds-fair/ * 筆記 * LLMs like ChatGPT are tools that reward power-users * The AI trust crisis: 如何說服用戶我們沒有拿他們資料去訓練? * Claude 3.5 Sonnet blog 備註中寫 根本沒有拿用戶資料去訓練,很棒 * Training on unlicensed scraped data was the original sin * prompt injection 還沒解決 * Don't publish slop! * Take accountability for the content that you produce * That's something LLMs will never be able to do * 我的 FB 分享 https://www.facebook.com/ihower/posts/10161196782253971 * Llamafile: bringing AI to the masses with fast CPU inference (Mozilla) * ✅ https://ihower.tw/watch/aief-2024-keynote-day1-llamafile/ * llamafile 可以讓 CPU 推論變快 30-500% faster (intel, AMD, arm) * Spreadsheets-are-all-you-need: Decoding the Decoder LLM without de code * ✅ https://ihower.tw/watch/aief-2024-spreadsheets-are-all-you-need/ * 用試算表 示範做一個 gpt-2 LLM,沒有 code,只有 excel function,這場演講把帶大家能夠一步一步理解 GPT 每一個階段在做什麼 * The Future of Knowledge Assistants (Jerry Liu) * ✅ https://ihower.tw/watch/aief-2024-llamaindex-the-future-of-knowledge-assistants/ * AI Engineering Without Borders, swyx * ✅ https://ihower.tw/watch/aief-2024-swyx-ai-engineering-without-borders/ * [ ] Convex Launch * [ ] Hasura Launch: Realtime Data Connectivity for AI * [ ] Hypermode Launch: LLMOps 產品 * [ ] Hyperspace: More Nodes Is All You Need: a peer-to-peer AI network - [ ] BotDojo Launch: Enhancing AI Assistants with Evaluations and Synthetic Data - [ ] Emergence Launch: AI Agents and the future enterprise - [ ] Second Order Effects ## Dev Tools * [ ] GitHub Next Explorations * Next 團隊直接向 CEO 回報,負責探索軟體開發。探索之後才轉給產品團隊開發。 * https://githubnext.com/ * Code-completions -> Task completions * https://github.blog/2024-01-17-a-developers-second-brain-reducing-complexity-through-partnership-with-ai/ * [ ] Embeddings are Stunting Agents: How Codeium Breaks Through the Ceiling for Retrieval * Our Context awareness 是強項 * 但這個 MEEF Benchmark 是 useful 嗎? 我們採用一個新指標 * Codeium 的 LLM Retriever 最厲害 * How do we do this? More Compute! - [ ] Cursor: Building the Human-AI Hybrid Engineer - [ ] The AI emperor has no DAUs: why most devs still don't use code AI - [ ] Code Generation and Maintenance at Scale - [ ] Self-Evolving Code with AI: Enhancing Quality and Security in CI ## GPUs & Inference - Scott Wu and the Making of Devin by Cognition AI - ✅ https://ihower.tw/watch/aief-2024-devin/ - 我的 FB 分享 https://www.facebook.com/ihower/posts/10161198639978971 - [ ] Covalent Launch: The GPU Cheatcode: Fine-tune 20 Llama Models in 5 Minutes - [ ] Compute & System Design for Next Generation Frontier Models - [ ] Breaking AI’s 1 Gigahertz Barrier, Groq - [ ] Accelerating Mixture of Experts Training With Rail-Optimized InfiniBand Networking in Crusoe Cloud - [ ] Unveiling the latest Gemma model advancements - [ ] Making Open Models 10x faster and better for Modern Application Innovation ## Day 2 Keynote * 5 practival steps to go from Software Developer to AI Engineer, AWS * ✅ https://ihower.tw/watch/aief-2024-from-software-developer-to-ai-engineer/ * Minecraft agent: https://github.com/build-on-aws/amazon-bedrock-minecraft-agent * 我的 FB 分享 https://www.facebook.com/ihower/posts/10161198197188971 * What's new from Anthropic and what's next, Anthropic * 沒有發表新東西,就是介紹 Claude 3.5 Sonnet 跟 Artifacts 功能而已,如果有在關注和使用 Claude 就不用看了 * LangChain Launch: Infrastructure for building reliable agents * ✅ https://ihower.tw/watch/aief-2024-langgraph/ * Langgraph Cloud 排隊處: https://bit.ly/langgraph-cloud-beta * 投影片 https://x.com/llama_index/status/1808164468708499482 - What We Learned From A Year of Building With LLMs - ✅ https://ihower.tw/watch/aief-2024-what-we-learned-from-a-year-of-building-with-llms - 講者們的同主題 blog 文章: https://applied-llms.org/#strategy-building-with-llms-without-getting-out-maneuvered * [ ] Unlocking Developer Productivity across CPU and GPU with MAX, Modular - [ ] Copilots Everywhere ## Multimodality - From Text to Vision to Voice: Exploring Multimodality with OpenAI - ✅ https://ihower.tw/watch/aief-2024-openai/ - 我的 FB 分享: https://www.facebook.com/ihower/videos/1202706250856646 我擷取了 live demo 其中一段是和 gpt-4o 對話和分享螢幕,進行 app 協作開發的過程,並配上中文字幕。 - live demo - gpt-4o 語音對話(改變語氣蠻有趣的)、視訊對話(用相機看書的其中一頁說話、分享螢幕給 chatbot 對話,開發 app 過程做 debug 對話) - 多模態影像解讀、Sora、voice engine model (voice clone 並換不同語言) - [ ] Substrate Launch: the API for modular AI - [ ] Moondream: how does a tiny vision model slap so hard? - [ ] The era of unbounded products: Designing for Multimodal I/O - [ ] State Space Models for Realtime Multimodal Intelligence - [ ] The Hierarchy of Needs for Training Dataset Development - [ ] The Multimodal Future of Education, Google ## Open Models - [ ] Decoding Mistral AI's Large Language Models (Mistral) - [ ] No more bad outputs with structured generation - [ ] Building SOTA Open Weights Tool Use: The Command R Family (Cohere) - [ ] Training Albatross: An Expert Finance LLM - [ ] Fixing bugs in Gemma, Llama & Phi-3 - [ ] Unveiling the latest Gemma model advancements (DeepMind) # 以下還沒釋出影片 ## RAG & LLM Frameworks - [ ] Navigating RAG Optimization with an Evaluation-Driven Compass - [ ] Pydantic is STILL all you need - 投影片 https://tome.app/fivesixseven/pydantic-is-still-all-you-need-clxgxz2xp0b0498210u77pznh ## Evals & LLM Ops - [ ] How to construct domain-specific LLM evaluation systems. - [ ] The GenAI Maturity Curve (or: You Probably Don’t Need Fine-Tuning) - [ ] Lessons from the Trenches: Building LLM Evals That Work IRL ⭐ - https://www.ai.engineer/worldsfair/2024/schedule/lessons-from-the-trenches-building-llm-evals-that-work-irl ## Agents - [ ] Building Reliable Agentic Systems, Factory - [ ] Using agents to build an agent company, crewAI ## AI Leadership - [ ] Understanding AI Stakes to Break Production Code - [ ] Real ROI: Lessons from Enterprises that Have already succeeded with LLMs at Scale - [ ] AI Platform Engineering - [ ] Hiring & Building an AI Engineering Team - [ ] E-Values: Evaluating the Values of AI - [ ] LLM Safeguards: Security, Privacy, Compliance, Anti-Hallucination ## Export Sessions - [ ] Building production RAG systems at scale (with 10s of millions users) Perplexity ⭐ - [ ] Building Time Series Foundation Models: The Journey to Success with Nixtla and Microsoft - [ ] Tame your LLMs with Rust - [ ] The future of RAG is agentic - [ ] Vector Search is NOT All You Need (mongodb) ## Workshops * Architecting and Testing Controllable Agents * https://www.ai.engineer/worldsfair/2024/schedule/architecting-and-testing-controllable-agents * https://x.com/RLanceMartin/status/1805632366213546424 2024/6/26 * https://docs.google.com/presentation/d/1QWkXi4DYjfw94eHcy9RMLqpQdJtS2C_kx_u7wAUvlZE/preview?slide=id.g273e7f400bc_0_0 * https://x.com/LangChainAI/status/1806712348348203037 2024/6/28 * 影片 https://www.youtube.com/watch?v=XiySC-d346E * LLMs for the working programmer. Become a 10x programming centaur today! * https://www.ai.engineer/worldsfair/2024/schedule/llms-for-the-working-programmer-become-a-10x-programming-centaur-today * https://github.com/go-go-golems/go-go-workshop * Low Level Technicals of LLMs * https://www.ai.engineer/worldsfair/2024/schedule/low-level-technicals-of-llms * https://docs.google.com/presentation/d/1O1zHLevzf2gQTlw5bs-l1raARSbIeK-buPTbjnDtx0Y/edit#slide=id.p * neo4j * https://github.com/neo4j-product-examples/genai-workshop/blob/main/genai-workshop.ipynb * mongodb * https://www.mongodb.com/developer/events/ai-agents-workshop-aiewf-2024/