RT Avi Chawla (@_avichawla)You're in an AI engineer interview at Apple.The interviewer asks:"Siri processes 25B requests/mo.How would you use this data to improve its speech recognition?"You: "Upload all voice notes from devices to iCloud and train a model"Interview over!Here's what you missed:Modern devices (like smartphones) host a ton of data that can be useful for ML models.To get some perspective, consider the number of images you have on your phone right now, the number of keystrokes you press daily, ...

If you found it insightful, reshare it with your network.Find me → @_avichawlaEvery day, I share tutorials and insights on DS, ML, LLMs, and RAGs. https://t.co/pf5AKPYepRAvi Chawla (@_avichawla):You're in an AI engineer interview at Apple.The interviewer asks:"Siri processes 25B requests/mo.How would you use this data to improve its speech recognition?"You: "Upload all voice notes from devices to iCloud and train a model"Interview over!Here's what you missed: https://t.co/uKIt7n1teK ...

Federated Learning - IBM provides a good video on Federated Learning [1]

Federated Learning Overview - Federated learning addresses the challenge of training ML models on private data residing on user devices [3] - It dispatches a model to end devices for training on private data and aggregates the trained models on a central server [4] - This approach reduces computation requirements on the server side by distributing most computation to user devices [3] Challenges in Federated Learning - Client devices have limited RAM and battery power, requiring efficient training methods [3] - Aggregating different models received from client devices to create a central model poses a challenge [3] - Privacy-sensitive datasets are often biased with personal likings and beliefs, leading to skewness in client data distribution [5] Data Privacy and Bias - Data on modern devices, such as images, messages, and voice notes, is mostly private [2][4] - The skewness in client data distribution, such as an overrepresentation of pet, car, or travel images, needs to be addressed [5] Application in Speech Recognition - Siri processes 25 billion (25B) requests per month, representing a large dataset for improving speech recognition [1] - The data on user devices can be leveraged to improve ML models [1]

AI Agent & Database Evolution - AI agents are challenging the traditional database model designed for human interaction [1] - The industry recognizes the need for databases to adapt to the requirements of AI agents, which differ significantly from human users [1] Agentic Postgres Features - TimescaleDB introduces Agentic Postgres, an agent-ready version of Postgres designed to address the challenges posed by AI agents [2] - Agentic Postgres enables instant database branching, facilitating parallel agent evaluations, safe experiments, migrations, and isolated testing with minimal cost and time [2] - It includes a built-in MCP server, offering schema guidance, best practices, and secure, structured access to Postgres for agents, aiding in informed migrations [3] - Hybrid search (vector search and BM25) is integrated, allowing agents to directly retrieve data within the database [3] - The database is memory-native, providing a persistent context for agent evolution [3] AI Agent Requirements - AI agents require the ability to branch endlessly, run multiple experiments concurrently, and operate within isolated, contextualized, and secure sandboxes [4]

Database & AI - AI agents are redefining the traditional role of databases [1] - Traditional databases were designed for human interaction, a model now challenged by AI agents [1] - AI agents require features like isolation, context, memory, and structured data [1] Postgres & AI - This is a significant moment for Postgres in the context of AI development [1] AI Agent Behavior - AI agents exhibit branching behavior [1] - AI agents conduct multiple experiments simultaneously [1]

Here's the usage of their MCP server.It's created based on 35 years of Postgres knowledge, and full access Postgres docs, all in a format that agents can easily process.You can try this live in Tiger Data's Free Tier here: https://t.co/vQVHBNnYHW. https://t.co/RK3TqdKaOj ...

Big moment for Postgres!AI agents broke the idea of what a database is supposed to do.Traditional databases were built for humans, and Agents broke that model.- They branch endlessly.- They run ten experiments at once.- They need isolation, context, memory, structured reasoning, and safe sandboxes.Letting agents touch production systems is terrifying because the old model of Postgres was never built for this kind of behavior.Agentic Postgres is an agent-ready version of Postgres by @TimescaleDB that solves ...

Security Concerns - The industry faces challenges in preventing adversarial attacks via prompts in LLMs [1] - OpenAI paid $500k in a Kaggle contest to find vulnerabilities in gpt-oss-20b [1] Model Evaluation - LLMs are evaluated against correctness [1]

LLM Security Concerns - The industry faces a common challenge: preventing adversarial attacks on LLMs via prompts [1] - OpenAI invested $500 thousand in a Kaggle contest to identify vulnerabilities in gpt-oss-20b [1] Key Players - OpenAI, Google, and Meta are all grappling with prompt-based adversarial attacks on LLMs [1]