Artificial Intelligence #267
Photo by Janik Butz

Artificial Intelligence #267

Hey, in this issue: the cultural divide between mathematics and AI; how to backdoor an AI model; a surveillance camera with AI anomaly detection; build your own GitHub Copilot; transformers without normalization; Mistral Small 3.1 SOTA vision model; and more.

The sponsor of this issue is Predibase.

Introducing Reinforcement Fine-Tuning (RFT): the best way to customize LLMs with just 10 labeled examples. Join Predibase on March 27 to learn how to beat commercial LLMs with RFT and open-source models, design reward functions with GRPO, and see our new OSS PyTorch-to-Triton model. Register now!

More than 850,000 subscribers are reading this newsletter. If you are building an AI or a data product or service, you can become a sponsor of one of the future newsletter issues and get your business featured in the newsletter. Feel free to reach out to [email protected] for more details on sponsorships.

Enjoy the newsletter? Please help us make it bigger and better by sharing it with colleagues and friends.

Raju Pawar

Turning Words into Influence & Impact Every brand has a story. Every Entrepreneur, Visionary Thinker has a message. But not everyone knows how to put it into words that matter. That’s where I come in!

16 小时前

Great service

回复
Vincent Granville

Co-Founder, BondingAI.io

18 小时前

A divide? Not in my case, it's a happy marriage: https://mltblog.com/3F3Y9Yd

回复
Ed Fowkes

Unlocking Growth for SMEs ?? TableNetwork Advisor ?? Speaker - Business and Property ?? Mentor ??

18 小时前

Sounds like a fantastic edition, Andriy! So much happening in AI right now. Looking forward to diving into these insights.

回复
Yefim Shuf, Ph.D.

Chief / Principal Architect | Technical Executive | Technology Leader | Inventor | Data Architect | MLOps BigData Analytics ML AI MLSecOps | NPS 100 | Chief Practice Officer & Client Partner Apex Team under CEO

21 小时前

* The following is worth reading and caught my attention -- "The Dark Side of Python’s pickle – How to Backdoor an AI Model" -- especially the "poisoned models" piece. I recall PMML exported models ( supported by SAS & SPSS and other tools) did NOT have such issues. i.e. safer. https://jchandra.com/posts/python-pickle/ * " The Hundred-Page Language Models Book" -- For the curious ones --- Scroll down to Chapters to pre-view the text of the book. Right below Buy the Book and above About the Author. Easy to miss unless you know where to look and what to look for. Reads well. Happy learning. https://thelmbook.com/ /

回复
Wayne Harburn

Next Generation Ai based Drone Systems

21 小时前

Awesome product

回复

要查看或添加评论,请登录

Andriy Burkov的更多文章

  • Artificial Intelligence #266

    Artificial Intelligence #266

    Hey, in this issue: companies are failing to convince staff of AI benefits; pioneers of reinforcement learning win the…

    9 条评论
  • Artificial Intelligence #266

    Artificial Intelligence #266

    Hey, in this issue: companies are failing to convince staff of AI benefits; pioneers of reinforcement learning win the…

    6 条评论
  • Artificial Intelligence #265

    Artificial Intelligence #265

    Hey, in this issue: How much energy will AI really consume?; how AI can achieve human-level intelligence; chatbots are…

    11 条评论
  • Artificial Intelligence #264

    Artificial Intelligence #264

    Hey, in this issue: emerging patterns in building GenAI products; the state of machine learning competitions;…

    15 条评论
  • Artificial Intelligence #264

    Artificial Intelligence #264

    Hey, in this issue: emerging patterns in building GenAI products; the state of machine learning competitions;…

    12 条评论
  • Artificial Intelligence #263

    Artificial Intelligence #263

    Hey, in this issue: the end of programming as we know it; your most important customer may be AI; the impact of…

    13 条评论
  • Artificial Intelligence #263

    Artificial Intelligence #263

    Hey, in this issue: the end of programming as we know it; your most important customer may be AI; the impact of…

    4 条评论
  • Artificial Intelligence #262

    Artificial Intelligence #262

    Hey, in this issue: a first major win for an AI copyright case in the US; your AI can’t see gorillas; AI-designed…

    13 条评论
  • Artificial Intelligence #262

    Artificial Intelligence #262

    Hey, in this issue: a first major win for an AI copyright case in the US; your AI can’t see gorillas; AI-designed…

    9 条评论
  • Artificial Intelligence #261

    Artificial Intelligence #261

    Hey, in this issue: How are researchers using AI?; no hype DeepSeek R1 reading list; RAG best practices; robotic…

    15 条评论