LLM Post-Training 101 + Prompt Engineering vs Context Engineering | AI & ML Monthly

Video Description

Welcome to machine learning & AI monthly for September 2025. This is the video version of the newsletter I write every month which covers the latest and greatest (but not always the latest) in the world of AI and ML. Read the issues online: - AI/ML Monthly September 2025 (this video) — https://zerotomastery.io/blog/ai-and-machine-learning-monthly-newsletter-september-2025/ - AI/ML Monthly July 2025 — https://zerotomastery.io/blog/ai-and-machine-learning-monthly-newsletter-july-2025/ My links: Download Nutrify (my startup) - https://nutrify.app Download KeepTrack (my other startup) - https://keeptrack.app Personal website - https://www.mrdbourke.com My ML blog - https://learnml.io Read my novel Charlie Walks - https://www.charliewalks.com Courses I teach: Learn Hugging Face - https://learnhuggingface.com Learn AI/ML (beginner-friendly course) - https://dbourke.link/ZTMMLcourse Learn TensorFlow - https://dbourke.link/ZTMTFcourse Learn PyTorch - https://dbourke.link/ZTMPyTorch Timestamps: 0:00 - Intro 0:30 - My Work 0:35 - Birthday article - https://www.mrdbourke.com/32/ 1:00 - LLM fine-tuning tutorial (work in progress) 1:05 - Object detection tutorial - https://www.learnhuggingface.com/notebooks/hugging_face_object_detection_tutorial 1:55 - Post-training 101 by Han Fang and Karthik Sankararaman - https://tokens-for-thoughts.notion.site/post-training-101 6:05 - Hugging Face TRL now supports native vision fine-tuning - https://huggingface.co/blog/trl-vlm-alignment 8:15 - Bonus: Nathan Lambert on SFT vs RL for small models 9:20 - Ethan Ding writes about Agents vs Clouds - https://ethanding.substack.com/p/agents-vs-clouds 12:30 - Modal releases notebooks - https://modal.com/blog/notebooks 14:35 - Case study: Hugging Face Is All You Need by Finegrain - https://huggingface.co/blog/finegrain/model-quality-hugging-face-all-you-need 18:06 - Case study: TensorZero finds 30x cost savings with fine-tuning - https://www.tensorzero.com/blog/distillation-programmatic-data-curation-smarter-llms-5-30x-cheaper-inference/ 21:30 - Thinking Machines: Making LLMs deterministic - https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/ 25:27 - Stephan Tulkens blog on tokenizers - https://stephantul.github.io/blog/uncasing/ 27:30 - Anthropic: Prompt Engineering vs Context Engineering - https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents 31:35 - Qwen3-Omni - https://huggingface.co/collections/Qwen/qwen3-omni-68d100a86cd0906843ceccbe 33:40 - Qwen3-Next - https://huggingface.co/collections/Qwen/qwen3-next-68c25fd6838e585db8eeea9d 35:30 - Qwen3Guard - https://huggingface.co/collections/Qwen/qwen3guard-68d2729abbfae4716f3343a1 36:45 - Qwen3-VL - https://huggingface.co/collections/Qwen/qwen3-vl-68d2a7c1b8a8afce4ebd2dbe 36:50 - Qwen3-VL blog post - https://qwen.ai/blog?id=99f0335c4ad9ff6153e517418d48535ab6d8afef&from=research.latest-advancements-list 36:55 - Qwen3-VL cookbooks - https://github.com/QwenLM/Qwen3-VL/tree/main/cookbooks 41:00 - Google DeepMind: EmbeddingGemma (308M parameter embedding model) - https://huggingface.co/blog/embeddinggemma 43:20 - Apple: MobileCLIP2 - https://github.com/apple/ml-mobileclip 47:10 - FineVision: 5TB of high-quality vision data - https://huggingface.co/spaces/HuggingFaceM4/FineVision 52:30 - Ettin encoder and decoder pairs outperform ModernBERT - https://huggingface.co/blog/ettin 56:06 - TinyLettuce Encoders for hallucination detection - https://huggingface.co/blog/adaamko/tinylettuce 58:05 - MinerU2.5 - https://huggingface.co/opendatalab/MinerU2.5-2509-1.2B 1:01:30 - POINTS-Reader - https://huggingface.co/tencent/POINTS-Reader 1:04:10 - PP-OCRv5 - https://huggingface.co/collections/PaddlePaddle/pp-ocrv5-684a5356aef5b4b1d7b85e4b 1:04:40 - Granite-Docling-258M - https://huggingface.co/ibm-granite/granite-docling-258M 1:05:10 - Z.ai: GLM-4.6 - https://huggingface.co/zai-org/GLM-4.6 1:06:10 - Mistral: Magistral with vision - https://huggingface.co/mistralai/Magistral-Small-2509 1:07:20 - Google DeepMind: Gemini Robotics 1.5 - https://developers.googleblog.com/en/building-the-next-generation-of-physical-agents-with-gemini-robotics-er-15/ 1:09:00 - François Chollet: Deep Learning with Python 3rd edition - https://deeplearningwithpython.io/ 1:09:35 - Hugging Face: trackio - https://huggingface.co/blog/trackio 1:10:10 - Google: Grounding with Google Maps in Gemini API - https://cloud.google.com/vertex-ai/generative-ai/docs/grounding/grounding-with-google-maps 1:10:25 - Anthropic: Claude Sonnet 4.5 - https://www.anthropic.com/news/claude-sonnet-4-5 1:10:35 - OpenAI: Sora 2 - https://openai.com/index/sora-2/ 1:11:20 - Video models are zero-shot learners - https://video-zero-shot.github.io 1:11:40 - Time series foundation models as few-shot learners - https://research.google/blog/time-series-foundation-models-can-be-few-shot-learners/ 1:13:01 - Talks 1:13:20 - Demis Hassabis at All In Summit - https://youtu.be/Kr3Sh2PKA8Y 1:14:20 - Outro