Benoit Lourdelet

I wrote a new post on the Sympathetic Ink blog to sum up The Deconstructed Database and what makes it composable. Learn more about the role of #Parquet, #Arrow, #DataFusion, Apache Iceberg, #Calcite and OpenLineage

64

4 commentaires

Today, we're diving into the cutting edge of Retrieval-Augmented Generation (RAG), a paradigm shift that's turbocharging LLMs with the vast knowledge of your BigQuery data. Think about it: 🤯 - LLMs are powerful, but they're only as good as the data they're trained on. - BigQuery is a data powerhouse, holding the keys to your unique insights. Enter RAG! By seamlessly integrating BigQuery with your LLMs, we unlock a whole new level of intelligence. 🧠 We dive deep into the mechanics, in this blog post we cover: - Understanding RAG's potential: Explore how it transforms LLMs from generic models to highly specialized experts in your domain. - Practical implementation: Learn how to seamlessly connect BigQuery and your LLM, leveraging Google's tools for effortless integration. - Real-world use cases: Discover how RAG unlocks powerful applications, from personalized recommendations to cutting-edge research. Click the link below to read the blog post and discover how to unleash the full potential of your LLMs: https://lnkd.in/e8jacn9a Also see the other blogs in the series and watch this space for more: 1 - Multimodal input to Gemini to BigQuery Schema 2 - Creating marketing campaigns using BigQuery and Gemini models https://lnkd.in/enJ5rxec Watch the video https://lnkd.in/ekzeNcs9 hashtag#RAG hashtag#BigQuery hashtag#LLMs hashtag#AI hashtag#DataAnalytics hashtag#GoogleCloud hashtag#Innovation hashtag#FutureofAI hashtag#LLM hashtag#Gemini hashtag#BigQuery hashtag#datawarehousing hashtag#Cloud hashtag#marketing hashtag#cooldemos hashtag#bigdata hashtag#lakehouse hashtag#genai hashtag#ai hashtag#artificialintelligence hashtag#datamodelling hashtag#google hashtag#dataengineering hashtag#analytics hashtag#ml hashtag#GenerativeAI hashtag#BigQuery hashtag#DataAnalytics hashtag#AIInnovation hashtag#GoogleCloud hashtag#DataScience hashtag#MachineLearning hashtag#gemini hashtag#vertexai hashtag#agents

9

Wondering how Looker’s Semantic Model is so unique? Watch the video below to uncover its value and how to create your own within seconds 💎 In summary, Looker’s Semantic Model (Develop menu) helps you IMPLEMENT AN END-TO-END DATA GOVERNANCE SYSTEM with four main value points, 🧰 Single source of truth: • Centralize your data metrics’ definitions using LookML, Google’s easy to learn coding language making SQL modular and reusable • Design a data model by defining the access rules and junctions between your database tables • Maximize data scalability, security, consistency, and maintenance efficiency 🎨 Self-service customization: • Build as many self service layer(s) or “Explores” as you want by choosing the metrics and tables you wish to display • Hide or show metrics dynamically based on user attributes to enable a personalized experienced so different teams can speak or use the data the way they prefer • Create quickstart templates or metric definitions to inspire and onboard colleagues 🏎️ Automatic SQL queries: • Translate AUTOMATICALLY and in REAL-TIME self-service field selections into optimized SQL queries • Ensure high-quality SQL capable of handling advanced use cases (e.g., fanouts, nested fields, custom caching, aggregate and derived tables…) and available for review in the “SQL” tab of your Explore • Overcome technical limitations related to groupings, filtering complexity, or metrics availability by creating any combination of metrics available within your model 🚀 Faster data modeling: • Leverage a fully GIT-integrated process with shared branches and advanced deployment options • Ramp up quickly with a dynamic Quick Help functionality, Google’s free trial support, or LookML Assistant suggesting code snippets from natural language prompts • Visualize your data model with free Looker extensions like the LookML Diagram or Data Dictionary, and troubleshoot raw data with the Content Validator and SQL Runner • Generate AUTOMATICALLY ∼70% of your first functional model when you connect your database(s) Interested to discuss it further? Contact your Google BI team (including Soufiane and Samy if you or your leadership are based in France) to schedule a call, discuss your use case, and explore a potential collaboration on a FREE TRIAL (limited spots, first come first serve!).

33

One of the best outcomes in trying to make Gen AI work @ scale for enterprises. " In the strenuous due course of creating a functional AI product and teaching the models, -Teams understand inefficiencies in existing systems that need operational fixes -Teams learn what and the how part of the systems to get better both in AI and Non AI fronts" As always teaching is a best form of refined learning, be it to humans or models.

58

2 commentaires

WSJ : For AI Giants, Smaller Is Sometimes Better Companies are turning their attention to less powerful models, hoping lower costs and solid performance will win more customers. ===== “ EJVM’s high-level view ==> cost of smaller models & scarcity of new high-quality data & small models are Faster & small models can be more Domain Specific to answer queries and do Fine-Tuning”. ===== This category of AI software—called small or medium language models—is trained on less data and often designed for specific tasks. The largest models, like OpenAI’s GPT-4, cost more than $100 million to develop and use more than one trillion parameters, a measurement of their size. Smaller models are often trained on narrower data sets—just on legal issues, for example—and can cost less than $10 million to train, using fewer than 10 billion parameters. The smaller models also use less computing power, and thus cost less, to respond to each query. Microsoft MSFT 1.47% has played up its family of small models named Phi, which Chief Executive Satya Nadella said are 1/100th the size of the free model behind OpenAI’s ChatGPT and perform many tasks nearly as well. The company = Microsoft also recently launched AI laptops that use dozens of AI models for search and image generation. The models require so little data that they can be run on a device and don’t require access to massive cloud-based supercomputers as ChatGPT does. Google—as well as AI startups Mistral, Anthropic and Cohere—have also released smaller models this year. Apple unveiled its own AI road map in June with plans to use small models so that it could run the software entirely on phones to make it faster and more secure. Even OpenAI, which has been at the vanguard of the large-model movement, recently released a version of its flagship model it says is cheaper to operate. A spokeswoman said the company is open to releasing smaller models in the future. For many tasks, like summarizing documents or generating images, large models can be overkill—the equivalent of driving a tank to pick up groceries. Because they use less computing power, small models can answer questions for as little as one-sixth the cost of large language models in many cases, said Yoav Shoham, co-founder of AI21 Labs, a Tel Aviv-based AI company. “If you’re doing hundreds of thousands or millions of answers, the economics don’t work” to use a large model, Shoham said. The key is focusing these smaller models on a set of data like internal communications, legal documents or sales numbers to perform specific tasks like writing emails—a process known as fine-tuning. That process allows small models to perform as effectively as a large model on those tasks at a fraction of the cost. Researchers attribute this to factors including a scarcity of high-quality, new data for training. More details in WSJ article URL below 👇

16

It gives us great pleasure to announce that Apache Spark in BigQuery is now officially available in Private Preview. This feature expands BigQuery’s processing capabilities by offering Apache Spark as an OSS processing engine under a unified control plane (admin, monitoring, billing, security, governance). It is the first step in delivering on the Unified Data Platform vision and was a key announcement at Next ‘24 #googlecloud #bigquery #spark

38

2 commentaires

Ever wondered how to leverage the power of Large Language Models (LLMs) in collaborative workflows? Look no further than AutoGen, Microsoft's open-source framework taking the tech world by storm! Imagine a team of AI agents, each with unique strengths, working together to solve problems. This is the magic of AutoGen. It empowers you to build LLM applications where agents converse, combining human expertise, powerful tools, and cutting-edge AI. 𝐇𝐞𝐫𝐞'𝐬 𝐰𝐡𝐚𝐭 𝐦𝐚𝐤𝐞𝐬 𝐀𝐮𝐭𝐨𝐆𝐞𝐧 𝐭𝐫𝐮𝐥𝐲 𝐚𝐦𝐚𝐳𝐢𝐧𝐠: > 𝐂𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐏𝐨𝐰𝐞𝐫𝐡𝐨𝐮𝐬𝐞𝐬: Create custom agents that can be LLMs themselves, use external tools, or even represent human input! This lets you build specialized agents for tasks like writing code, executing actions, validating results, and seamlessly integrating human feedback. > 𝐏𝐫𝐨𝐠𝐫𝐚𝐦𝐦𝐢𝐧𝐠 𝐂𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧𝐬: Forget complex coding! AutoGen simplifies building intricate LLM applications by letting you program agent interactions naturally. Define agent roles and capabilities, then design their conversations using plain language or code. > 𝐅𝐥𝐞𝐱𝐢𝐛𝐢𝐥𝐢𝐭𝐲 𝐑𝐞𝐢𝐠𝐧𝐬 𝐒𝐮𝐩𝐫𝐞𝐦𝐞: AutoGen adapts to your needs. It supports various conversation patterns, agent behaviors, and human involvement levels. Whether you need static or dynamic interactions, single-turn or multi-turn dialogues, AutoGen empowers you to create the perfect workflow. AutoGen's potential stretches far and wide. From tackling complex math and coding problems to providing insightful answers and optimizing operations, it's a game-changer across industries. What do you think of this? #innovation #technology #future #management #startups

21

1 commentaire

Great blog from Andreas that discusses architecture and performance comparison between Amazon RDS for PostgreSQL Single-AZ DB instance, Multi-AZ DB instance, and Multi-AZ DB Cluster deployments! #AmazonRDS #rds #performance #availability #multi_az_db_cluster https://lnkd.in/gZpVcdnU

40

Another day, another LLM model announcement. Snowflake have announced #SnowflakeArctic Key Points: - 480 billion parameter model - Truly open model - Apache 2 licensed with access to weights, code, data recipes and research insights (now that is more like it !) - Research is shared on Arctic 'cookbook' 👉 https://lnkd.in/ePmgeV_d "we see a consistent pattern in AI needs and use cases from our enterprise customers" "Enterprises want to use LLMs to build conversational SQL data copilots, code copilots and RAG chat bots." #ai #enterpriseai https://lnkd.in/ePiwZU-B

2

🏃‍♂️💨 We’re taking fast on tour! Come and see us in person at #RedisReleased, our exclusive customer events in 🇮🇳 Bengaluru on August 23 and 🇸🇬 Singapore on August 29 (and coming to 🇺🇸 New York and 🇬🇧 London this fall). Connect with the #Redis community and see how some of the most innovative companies in the world are using Redis to build the future, fast. Here’s what we’ll be covering: 💡 Product announcements you won’t want to miss 🛠 Hands-on workshops about building GenAI apps with Redis 🤔 What’s next for the world’s fastest data platform Register here: https://redis.io/released/

14

2 commentaires

New feature: "A data export button for Network, Exchange, and Facilities - click it and select what you want to download as structured data!"

8

“Meta AI chief says large language models will not reach human intelligence.” Financial Times (link in comments). I rarely step into the AGI debate - a lack of definition and too many unknowns - but this makes sense to me. Yann LeCun said to the FT: “[LLMs have] very limited understanding of logic . . . do not understand the physical world, do not have persistent memory, cannot reason in any reasonable definition of the term and cannot plan . . . hierarchically”. The area I’m interested in - culture creation - is socially constructed. To make it, you need to be living in the real world and having human relationships. So, the same reason we don’t (yet) have AGI, may be the reason we still need a human in the loop to make any meaningful art with AI. #agi #ArtificialGeneralIntelligence #CreativeReality

5

3 commentaires

Elasticsearch open inference API adds support for OpenAI chat completions. Learn more in this blog post: https://gag.gl/hEDnXw #ElasticSearchLabs #OpenAI

5

Have you missed the Data+AI Summit ? Don't worry we got you covered. Here is the Gen AI DAIS recap presented by my amazing colleagues Lara Rachidi and Maria Zervou. Next newsletter will be about Data Engineering recap 😎 #Databricks #DataAISummit

10

In the rapidly evolving landscape of Large Language Models (LLMs), traditional benchmarks like the MMLU (Massive Multitask Language Understanding) have been the go-to for evaluating AI capabilities. However, a growing body of research suggests that relying solely on these scores to select LLMs for specific applications could be misleading and costly. 🔍 𝐖𝐡𝐲 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐬 𝐌𝐢𝐠𝐡𝐭 𝐌𝐢𝐬𝐥𝐞𝐚𝐝: A recent study, "When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards," underscores a critical vulnerability in the standard benchmarking approach. This research demonstrates how minor methodological changes can significantly shift LLM rankings by up to eight positions on leaderboards. Such volatility raises questions about the stability and reliability of benchmark-based assessments. 👩‍💻 𝐓𝐡𝐞 𝐋𝐢𝐦𝐢𝐭𝐚𝐭𝐢𝐨𝐧𝐬 𝐨𝐟 𝐌𝐂𝐐𝐬: Leaderboards often rely on multiple-choice questions (MCQs) to gauge model performance. While MCQs offer a straightforward and quantifiable way to test certain model capabilities, they fall short in measuring comprehensive performance. The research highlights that models might perform well on MCQs by recognizing patterns rather than genuinely understanding content, which doesn’t necessarily translate to effectiveness in practical applications. 🔄 𝐓𝐡𝐞 𝐬𝐭𝐮𝐝𝐲 𝐜𝐚𝐭𝐞𝐠𝐨𝐫𝐢𝐳𝐞𝐬 𝐚𝐝𝐣𝐮𝐬𝐭𝐦𝐞𝐧𝐭𝐬 𝐢𝐧𝐭𝐨 𝐬𝐞𝐧𝐬𝐢𝐭𝐢𝐯𝐞 𝐚𝐧𝐝 𝐮𝐧𝐬𝐞𝐧𝐬𝐢𝐭𝐢𝐯𝐞 𝐩𝐞𝐫𝐭𝐮𝐫𝐛𝐚𝐭𝐢𝐨𝐧𝐬 𝐭𝐨 𝐞𝐱𝐩𝐥𝐨𝐫𝐞 𝐭𝐡𝐞𝐢𝐫 𝐢𝐦𝐩𝐚𝐜𝐭 𝐨𝐧 𝐦𝐨𝐝𝐞𝐥 𝐫𝐚𝐧𝐤𝐢𝐧𝐠𝐬: - Sensitive Perturbations: Modifications like shuffling choice order, changing symbols, and alternative scoring drastically alter model standings, pointing to a fragility in how performance is currently measured. - Unsensitive Perturbations: Adjustments that tweak the relevance of in-context examples show minor performance variations and help mitigate bias, suggesting a need for more nuanced testing methodologies. 𝐅𝐨𝐫 𝐨𝐫𝐠𝐚𝐧𝐢𝐳𝐚𝐭𝐢𝐨𝐧𝐬 𝐥𝐨𝐨𝐤𝐢𝐧𝐠 𝐭𝐨 𝐢𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐞 𝐋𝐋𝐌𝐬 𝐢𝐧𝐭𝐨 𝐭𝐡𝐞𝐢𝐫 𝐨𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬, 𝐭𝐡𝐢𝐬 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐬𝐞𝐫𝐯𝐞𝐬 𝐚𝐬 𝐚 𝐜𝐚𝐮𝐭𝐢𝐨𝐧𝐚𝐫𝐲 𝐭𝐚𝐥𝐞: - Custom Evaluation: Assess LLM performance using your specific data and scenarios rather than relying exclusively on generalized benchmarks. This approach ensures the model aligns with your unique requirements. - Holistic Assessment: Consider a broader range of evaluations that go beyond MCQs to include tests that measure real-world task performance and contextual understanding. - Continuous Benchmarking: As LLM technology evolves, so should the methods we use to evaluate it. Businesses must stay updated on the latest research and adapt their evaluation strategies accordingly. Do you think traditional benchmarks suffice, or is there a need for innovation in evaluation methods? #ai #genai #innovation #technology #llms

21

2 commentaires

The Architecture of Apache Druid #apachedruid #olap #history #architecture #storage #distributedsystems https://lnkd.in/egqai_ui

1