I spent the week tracking basically every major AI announcement, and it was honestly hard to keep up. In this March AI news round-up, I break down OpenAI’s surprise GPT-5.4 launch (standard, Thinking, and Pro), the DeepSeek V4 rumors that didn’t actually materialize, and why Anthropic’s Claude updates are pushing a lot of people to jump ship from ChatGPT. We also hit the business side: Cursor quietly crossing $2B in annualized revenue, and Claude shooting to #1 on the Apple App Store after the Pentagon “rejection” drama. My biggest takeaway: press releases and benchmarks don’t matter if the model doesn’t perform on real work. I tested GPT-5.4 the day it dropped and came away thinking it’s overrated—especially compared to Sonnet 4.6 inside Claude Co-work. OpenAI talked up spreadsheets, so I tried a real spreadsheet cleanup workflow and it was subpar, while Sonnet 4.6 did a fantastic job and kept clear notes on what changed. On the Claude side, memory is now on the free plan and importing ChatGPT history is frictionless, which is a smart move if you’re trying to win users during a messy news cycle. We wrap with the largest anti-AI protest yet in London’s tech hub and what that sentiment gets right (like “stop the slop”).