Where AI stands in 2026


Hi Reader, happy new year! 🎉

I wanted to share with you the three most important articles I found that look back at AI progress in 2025 and look forward at what is coming in 2026 and beyond.

I’ve extracted the key points from each article, but if you have the time and interest, I’d encourage you to read the full articles!


💠 The Shape of AI: Jaggedness, Bottlenecks and Salients​

By Ethan Mollick

  • “Jaggedness” describes the uneven abilities of AI: It’s superhuman in some areas and far below human level in other areas.
  • The inability of LLMs (Large Language Models) to immediately create new memories and permanently learn from them is an important source of “bottlenecks” to AI progress.
  • “Jaggedness creates bottlenecks, and bottlenecks mean that even very smart AI cannot easily substitute for humans. At least not yet.”
  • Sometimes, when a single “jagged weakness” is fixed, “the entire system can jump forward.”
  • “If you want to understand where AI is headed, don’t watch the benchmarks. Watch the bottlenecks. When one breaks, everything behind it comes flooding through.”

🙈 Presenting the Case That the Future Will Be Unrecognizable​

By Steve Newman

  • When AI becomes more adaptable than humans, the world may transform to such a great extent that it will become unrecognizable to present-day humans.
  • This level of AI could be more impactful than “agriculture, the Industrial Revolution, and the atomic bomb”, and as impactful as “the emergence of Homo Sapiens as the dominant species on Earth.”
  • “I would not be surprised to see AI to cross the critical threshold…in a matter of decades, and the Unrecognizable Age to follow fairly soon thereafter.”

📆 2025 LLM Year in Review​

By Andrej Karpathy

  • LLMs developed reasoning-like strategies by being graded on objective problem solving rather than subjective human feedback.
  • LLM intelligence is jagged, but so is human intelligence.
  • Benchmarks of LLM performance are no longer trustworthy.
  • “Not only does vibe coding empower regular people to approach programming, it empowers trained professionals to write a lot more (vibe coded) software that would otherwise never be written.”
  • Google’s Nano Banana image generator is an “early hint” of what the GUI (Graphical User Interface) for LLMs will look like, since “people love to consume information visually” and thus “LLMs should speak to us in our favored format.”

🤔 In case you were wondering…

Do I use AI to write these emails? No.

If my goal was to send you a newsletter every week, regardless of whether I had something “important” to say, then AI would be a great way to achieve that!

Instead, I’ve decided to only email you when I actually have something worthwhile to say. And in those cases, you deserve to hear directly from me!


I hope your year is off to a great start! I’ll see you in the next newsletter 😄

- Kevin

P.S. Many people have asked when my Machine Learning book will be ready...

Unfortunately, due to some challenging personal circumstances, I was not able to finalize the book in December as I had promised. 🤦‍♂️ However, I still hope to publish it very, very soon! 🤞

In the meantime, you can download the first three chapters for free!

Learn Artificial Intelligence from Data School 🤖

Join 25,000+ intelligent readers and receive AI tips every Tuesday!

Read more from Learn Artificial Intelligence from Data School 🤖

Hi Reader, I'm excited to announce an "early Black Friday" sale of Data School's All-Access Pass 🎟 It's simple: You pay $99, and you get access to ALL of my courses for a year. The sale starts Monday, November 23, and ends on Cyber Monday, November 30. More info is below. Let me know if you have any questions! 💬 - Kevin All-Access Pass FAQ Which courses will I be able to access? While your subscription is active, you'll have access to ALL of Data School's paid courses: Build an AI chatbot...

Hi Reader, For 12 years, I've been teaching scikit-learn. It's the ideal tool for Machine Learning in Python, given its broad functionality, community support, and high-quality documentation. But there's a HUGE difference between documentation and an effective user manual. That's why I published Master Machine Learning with scikit-learn: I wanted to provide battle-tested guidance for achieving superior Machine Learning results. (Side note: Amazon US is currently offering a 34% discount on my...

Hi Reader, I hope you're having a great September! 🍂 Two quick announcements for you: 1️⃣ I'm working on two mystery courses right now: The first is an ML course, and the second has nothing to do with Data Science! 🤯 Can you guess what the second course is about? 2️⃣ On Monday, I'm raising the price of my Master Machine Learning book from $19 to $29. But until Monday, you can still snag a paperback copy for $19! The ebook will continue to be priced at $19, and you can still read the book...