AI Note-Takers Are Taking Over Meetings

AI-powered note-taking tools—now built into Zoom, Microsoft Teams, Google Meet, Otter.ai, and ChatGPT—are increasingly showing up to meetings in place of humans. In one example, six people attended a meeting while ten AI bots took notes.

I can’t wait to have access to some of these bots at work. Can these bots replace in-person learning? In a related field, what does that mean for universities and traditional lectures (and their astronomical tuition costs)? Parents are not going to pay hundreds of thousands in college tuition for their kids to send an AI bot to class to learn for them!

I also suspect most casual users of these bots haven’t thought about the data privacy and security implications for themselves, their families and their employers.

Spending Inference Time

Kevin Lu explores how to optimize the “inference” step in AI models—the moment when the model responds to your input. He argues that smart design choices during inference often matter more than model size. Lu explains how generating multiple candidate outputs in parallel (like Midjourney does with images) can improve quality with minimal speed loss. The key is balancing latency, cost, and output quality—especially in real-world applications where speed and cost drive user satisfaction.

What is the right trade-off between inference compute and inference time? Would we rather get faster or better answers from our AI assistants? The answer can’t always be both.

Microsoft to Cut 9,000 Jobs as Chatbots Take Over

Microsoft will lay off 9,000 employees—about 4% of its workforce—as it shifts more aggressively toward AI. (Although I suspect some but not all of this cut is just vanilla corporate restructuring.) CEO Satya Nadella revealed that AI now contributes to up to 30% of Microsoft’s codebase. This fits in with what I’m seeing with various programming Copilot tools and my experience at work. If anything, we should see AI contributing much higher percentages of other companies codebases (ie less complex codebases and where we are less concerned with security vulnerabilities). This plays into the general theme of elimination of entry level white collar jobs. Those are the easiest jobs to be replaced with AI and very often our entry level hires don’t contribute to the team or company until they have gained a few years of experience.