Hi Everyone!
Each week I send out a succinct article highlighting events and sharing resources AI engineers should know about. Subscribe if you want them in your inbox. Please support the authors of the below resources. A huge thanks to all my supporters! If you would like, you can support my writing for just $2/mo.
If you want to learn AI/machine learning, I created a roadmap to do it all entirely for free. Check it out here. Enjoy this week’s resources!
Always be (machine) learning,
Logan
Events you should know about
YCombinator is hosting its first-ever AI Startup School. They will be gathering the top 2,000 computer science students and graduates. Attendees will learn from leading AI experts. The conference is free and travel costs will be covered up to $500.
The Paris AI Summit focused on discussing AI’s challenges and opportunities at the international level. The US and UK refused to sign a declaration for inclusive and sustainable AI. This has economic and social implications for the future of AI.
Elon Musk made a bid for OpenAI. Responding to it, OpenAI leaders said OpenAI is not for sale. Musk claims this is to make OpenAI open again while OpenAI leaders claim Musk was the first to suggest OpenAI be made for-profit and closed in the first place.
Yann LeCun presented at the Paris AI Summit and made the claim that LLMs aren’t the future. His claim highlights the potential need for another breakthrough to overcome the limitations of the transformer architecture. LeCun says anyone wanting to get into AI should focus on that instead.
I stated my excitement for the first feature-length AI film on Twitter. I got a lot of backlash that made me realize just how little most people understand about AI and how it will impact society. The most surprising thing to me was the inability for many people to think about what AI will be able to accomplish instead of what it can accomplish. Many statements were made as if AI generation won’t improve with time. Please learn what you can about AI to understand how it’ll impact you.
What you missed last week
Resources you should read
Agents Simplified: What we mean in the context of AI
AI agents are semi- or fully-autonomous systems that use large language models (LLMs) for decision-making and problem-solving. They can access various tools and have memory capabilities, allowing them to interact with the real world and execute tasks independently. Building AI agents involves integrating multiple components, including prompts, tools, and reasoning processes.
The Shape of AI to Come! Yann LeCun at AI Action Summit 2025
Yann LeCun discusses the future of artificial intelligence at the AI Action Summit 2025. He highlights the advancements in AI technologies and their potential impact on society. LeCun emphasizes the importance of ethical considerations as AI continues to evolve.
Sharing the latest Model Spec | OpenAI
OpenAI has updated its Model Spec to enhance AI behavior, focusing on customizability, transparency, and user safety. The new version is now in the public domain, allowing developers and researchers to freely use and adapt it. OpenAI aims to continuously improve AI alignment and invites community feedback to guide future updates.
Jeff Dean & Noam Shazeer – 25 years at Google: from PageRank to AGI
By
Jeff Dean and Noam Shazeer discuss the rapid advancements in AI and its potential to transform productivity, especially in coding and problem-solving. They emphasize the importance of innovative hardware and continual learning in making AI systems more efficient and accessible. The conversation highlights the exciting future of AI, predicting significant improvements in capabilities and development processes.
Deep Research, information vs. insight, and the nature of science
By
OpenAI’s Deep Research is revolutionizing scientific progress by enhancing knowledge workers' ability to access and summarize existing information. While AI tools can accelerate research, they may not lead to novel discoveries, prompting a reevaluation of how science operates. The integration of AI into scientific practices could change the dynamics of knowledge creation and challenge traditional paradigms in the scientific community.
Foundations of Large Language Models
The book "Foundations of Large Language Models" focuses on essential concepts rather than advanced technologies. It covers four key areas: pre-training, generative models, prompting techniques, and alignment methods. Aimed at students and professionals in natural language processing, it serves as a valuable reference for understanding large language models.
How to Learn about AI for Non Technical Users
By
Non-technical users can learn about AI without needing to build systems from scratch, which is often unnecessary and time-consuming. The article offers a more efficient learning guide that focuses on practical insights rather than deep technical knowledge. It aims to help users interface better with industry trends and developers while avoiding the pitfalls of AI hype.
The Most Beautiful Algorithms Ever Designed
By
Some algorithms in computer science are considered beautiful due to their elegance, cleverness, and broad impact. The author highlights a few examples, including Binary Search and QuickSort, which showcase innovative problem-solving techniques. Future posts will delve deeper into these algorithms, exploring their unique insights and significance.
How Will AI Impact Law Firms?
By
AI is transforming law firms by improving efficiency and productivity, allowing lawyers to handle more cases with higher quality work. Many lawyers do not fear job loss due to AI, seeing it as a tool to enhance their capabilities rather than replace them. However, concerns exist about over-reliance on AI, where individuals might substitute their critical thinking with automated solutions.
LLM Visualization
The walkthrough explains how the nano-gpt model sorts a sequence of letters, specifically "C B A B B C" into "ABBBCC". Each letter is treated as a token, assigned an index, and converted into a vector before being processed through transformer layers. The model predicts the next token based on probabilities, allowing for continuous feedback and improvement.
