Mircea Trifan

RLHF shouldn't feel like a wall of jargon. In The RLHF Book, @natolambert makes reinforcement learning from human feedback clear, practical, & grounded in the bigger picture — from supervised learning to HCI to economics & philosophy. If you work with modern AI, this is foundational. 50% off through tomorrow (plus other bestsellers): hubs.la/Q042Zx9L0 Plus, BOGO on single products & 20% off Manning Online Pro Annual.

5 practical AI books worth reading ✯ 𝐀𝐈 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 by Chip Huyen ➫ Comprehensive guide for building AI applications with foundation models, covering system design, deployment, and scaling. ✯ 𝐓𝐡𝐞 𝐋𝐋𝐌 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫'𝐬 𝐇𝐚𝐧𝐝𝐛𝐨𝐨𝐤 by Paul Iusztin and Maxime Labonne ➫ Practical manual for LLM development covering data engineering, fine-tuning, RAG pipelines, and production deployment with AWS implementations and downloadable code. ✯ 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐋𝐋𝐌𝐬 𝐟𝐨𝐫 𝐏𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 by Louis-François Bouchard and Louie Peters ➫ Focuses on deploying LLMs in production environments using prompt engineering, fine-tuning, and RAG techniques for scalable AI applications. ✯ 𝐁𝐮𝐢𝐥𝐝 𝐚 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥 (𝐟𝐫𝐨𝐦 𝐒𝐜𝐫𝐚𝐭𝐜𝐡) by Sebastian Raschka, PhD ➫ Teaches building transformer-based LLMs from the ground up using PyTorch with examples designed to run on ordinary laptops. ✯ 𝐇𝐚𝐧𝐝𝐬-𝐎𝐧 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 by Jay Alammar and Maarten Grootendorst ➫ Features over 275 custom illustrations and working code examples with a visual approach to make LLM concepts accessible. Which other books would you like to add?

🚨 Prompt engineering is officially outdated. Anthropic just released the real playbook for building AI agents that actually work. It’s a 30+ page deep dive called The Complete Guide to Building Skills for Claude and it quietly shifts the conversation from “prompt engineering” to real execution design. Here’s the big idea: A Skill isn’t just a prompt. It’s a structured system. You package instructions inside a SKILL.md file, optionally add scripts, references, and assets, and teach Claude a repeatable workflow once instead of re-explaining it every chat. But the real unlock is something they call progressive disclosure. Instead of dumping everything into context: • A lightweight YAML frontmatter tells Claude when to use the skill • Full instructions load only when relevant • Extra files are accessed only if needed Less context bloat. More precision. They also introduce a powerful analogy: MCP gives Claude the kitchen. Skills give it the recipe. Without skills: users connect tools and don’t know what to do next. With skills: workflows trigger automatically, best practices are embedded, API calls become consistent. They outline 3 major patterns: 1) Document & asset creation 2) Workflow automation 3) MCP enhancement And they emphasize something most builders ignore: testing. Trigger accuracy. Tool call efficiency. Failure rate. Token usage. This isn’t about clever wording. It’s about designing an execution layer on top of LLMs. Skills work across Claude.ai, Claude Code, and the API. Build once, deploy everywhere. The era of “just write a better prompt” is ending. Anthropic just handed everyone a blueprint for turning chat into infrastructure. Download the guide here: resources.anthropic.com/hubfs/The-Comp…

Google just dropped another banger 🤯 It’s called PaperBanana, a new tool that generates publication-ready academic illustrations directly from your methodology text. 100% Free.

Extracts triplets from text into interactive graphs github.com/robert-mcdermo…

Hugging Face just made fine-tuning 10x easier. they released HF skills that you can plug into Claude Code, Codex, and Gemini to: > write training scripts > submit jobs to cloud GPUs > monitor the progress > push the models to HF hubs it's not just for fine-tuning, but also model evaluation, paper publishing, and dataset curation. building a solid HF profile can get you very far in AI space, and there's no excuses anymore (unless for the 💸) you can read the guide on how to use this skill on this blog: huggingface.co/blog/hf-skills… @huggingface

We used Claude Code to train open LLMs. Check out the tutorial. basically, we plugged HF skills into claude code and it was able to train LLMs end-to-end. Best thing, this works on all major coding agents: Codex, Cursor, and Gemini CLI. - You tell the agent to fine-tune a model on a dataset. You can define the dataset or let it search. - Agent picks hardware based on model size and checks dataset. - Job runs on cloud gpus. Either main run, or test run. - Agent share real-time progress dashboard with Trackio. - Checkpoints are pushed to HF. Take it for a whirl now in your fav coding agent.

#AI and #MachineLearning for Coders — Programmer's Guide to Artificial Intelligence: amzn.to/2FRYRtr ★★★★★ —————— #DataScientist #DataScience #DeepLearning #NeuralNetworks #ComputerVision #Coding

OpenSIR introduces an LLM self-play framework for open-ended reasoning It empowers models to generate & solve novel problems without external supervision, achieving significant improvements on benchmarks like GSM8K and College Math through adaptive difficulty & diverse exploration.

The most effective AI Agents are built on these core ideas. It's what powers Claude Code. It's referred to as the Claude Agent SDK Loop, which is an agent framework to build all kinds of AI agents. (bookmark it) The loop involves three steps: Gathering Context: Use subagents (parallelize them for task efficiency when possible), compact/maintain context, and leverage agentic/semantic search for retrieving relevant context for the AI agent. Hybrid search approaches work really well for domains like agentic coding. Taking Action: Leverage tools, prebuilt MCP servers, bash/scripts (Skills have made it a lot easier), and generate code to take action and retrieve important feedback/context for the AI agent. Turns out you can also enhance MCP and token usage through code execution and routing, similar to how LLM routing increases efficiency in AI Agents. Verifying Output: You can define rules to verify outputs, enable visual feedback (this becomes increasingly important in multimodal problems), and consider LLM-as-a-Judge to verify quality based on fuzzy rules. Some problems will require visual cues and other forms of input to perform well. Don't overcomplicate the workflow (eg, use computer-using agents when a simple Skill with clever scripts will do). This is a clean, flexible, and solid framework for how to build and work with AI agents in all kinds of domains.

🚨 Anthropic just solved the problem every AI agent engineer’s been screaming about for a year. Every agent today burns tokens like fuel every tool call, every definition, every intermediate result jammed into context. Now Anthropic’s introducing the fix: code execution with MCP. Instead of calling tools directly, agents now write code to call them and it changes everything. → 98.7% fewer tokens → 10x faster task completion → No context overload → No data leakage Think of it like this: Old agents talk about what to do. New agents just code and do it. Cloudflare calls it “Code Mode.” Anthropic just made it real. This is the turning point. The next generation of AI agents won’t prompt tools they’ll build with them.

#MachineLearning Solutions Architect Handbook — Practical Strategies and Best Practices in the #ML Lifecycle, System Design, #MLOps, and Generative AI: amzn.to/4bx8t6b v/ @PacktDataML —————— #DataScience #DataScientist #AI #GenAI #LLMs #LLMOps

I'm thrilled to share that Build a Multi-Agent System (From Scratch) for @ManningBooks has already sold over 400 copies in early access (@ManningMEAP)! Learn how to build LLM agents and MAS from scratch using open-source LLMs with @ollama 🦙. If you're interested in grabbing a copy, it's Manning's Deal of the Day today: mng.bz/WrEx

she actually summarized everything you must know from the “AI Engineering” book in 76 minutes. if you don’t got the time to read the book, you need to watch this. foundational models, evaluation, prompt engineering, RAG, memory, fine-tuning and many more. great starting point.

i’ve had the opportunity to be one of the early reviewers of this book by @aureliengeron last year and was gifted five books of my choice from O’Reilly 😄 this is an amazing introduction to ML for beginners. i’m so happy there’s finally a PyTorch option, TF is so dead ☠️

𝗛𝗼𝘄 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 If you want to learn about Software Architecture, you can start with the book: 𝗙𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀 𝗼𝗳 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 (amzn.to/4nfBWHd) by Mark Richards and Neal Ford. Here is 𝘁𝗵𝗲 𝗽𝗼𝗱𝗰𝗮𝘀𝘁 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝗯𝗼𝗼𝗸 by Rebecca Parsons (35mins), who is Thoughtworks CTO (thoughtworks.com/insights/podca…). Also, you can check the new edition of 𝗛𝗲𝗮𝗱 𝗙𝗶𝗿𝘀𝘁 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 (amzn.to/4hhaG9L), by Raju Gandhi with Mark and Neal. Then, if you want to expand you horizons above this, check 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗳𝗼𝗿 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀 by Simon Brown - leanpub.com/software-archi… (focus on modeling and documentation) and 𝗧𝗮𝗹𝗸𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗧𝗲𝗰𝗵 𝗟𝗲𝗮𝗱𝘀: 𝗙𝗿𝗼𝗺 𝗡𝗼𝘃𝗶𝗰𝗲𝘀 𝘁𝗼 𝗣𝗿𝗮𝗰𝘁𝗶𝘁𝗶𝗼𝗻𝗲𝗿𝘀 by Patrick Kua (real-life stories from projects by tech leads - amzn.to/46XURkN). In addition, the next step could be the following ones: - 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗶𝗻 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲, 𝟰𝘁𝗵 𝗘𝗱𝗶𝘁𝗶𝗼𝗻 (the bible of Software Architecture - amzn.to/4hnrIDm) - 𝗖𝗹𝗲𝗮𝗻 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 by Robert C. Martin (amzn.to/4ouutoE) - 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝘆 by Dave Farley (amzn.to/43ifhCI) - 𝗝𝘂𝘀𝘁 𝗘𝗻𝗼𝘂𝗴𝗵 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 by George H. Fairbanks (amzn.to/3L1rjtT) - 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 𝗜𝘁! by Michael Nygard (amzn.to/3KX4LKX) - 𝗥𝗶𝗴𝗵𝘁𝗶𝗻𝗴 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 by Löwy Juval (amzn.to/3KSzuc9) - 𝗧𝗵𝗲 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁 𝗘𝗹𝗲𝘃𝗮𝘁𝗼𝗿 by Gregor Hohpe (amzn.to/4ouUhRC) In addition, check 𝘁𝗵𝗲 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗵𝗼𝘄 𝘁𝗼 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁 (roles, responsibilities, and skills) in the image below.