Jensen Huang's Two-Hour Keynote and My Still-Empty GPU Cart
Nvidia's GTC keynote promised AI factories and a chip named after an astronomer, but months after launch I still can't buy an RTX 5090 anywhere near its sticker price.
The MacBook Air Finally Has Enough RAM
Apple quietly doubled the base RAM on the new M4 MacBook Air to 16GB without raising the price, ending years of shipping 8GB base models. A small, overdue fix worth noticing.
The Site Went Down and Somehow It Was Ukraine's Fault
X went down in three waves yesterday and Musk blamed Ukraine within the hour. Here's why I'm not buying the instant, evidence-free version of events.
The New Mac Studio's Top Chip Is Two Generations Behind
Apple's new Mac Studio buries an M3 Ultra chip above the newer M4 Max, and the naming makes way less sense than it should.
Monster Hunter Wilds Runs Like Garbage on My Perfectly Fine PC
Monster Hunter Wilds is a massive hit, but a decent mid-range PC still can't hold 60fps without frame generation propping it up.
Firefox Got Its First-Ever Terms of Service and Immediately Regretted It
Mozilla gave Firefox its first-ever terms of use on February 26th, the wording spooked a lot of longtime users, and the hasty rewrite two days later didn't really fix the trust problem.
I Let an AI Loose in My Terminal for an Afternoon
Anthropic dropped Claude 3.7 Sonnet and a new terminal coding tool called Claude Code today, so I installed it and pointed it at my worst Python script.
Majorana 1 and My Deja Vu Problem
Microsoft's Majorana 1 quantum chip announcement gave me deja vu, given how the last time they claimed to find Majorana particles, it ended in a retraction.
The $9.74 Billion Joke
Elon Musk's $97.4 billion bid for OpenAI's nonprofit got rejected, and Sam Altman's snarky $9.74 billion counter-joke about buying Twitter is the part actually worth talking about.
So Now It's Called "Vibe Coding"
Karpathy coined "vibe coding" a couple weeks ago and it's everywhere now. Here's how I actually use Cursor and Claude day to day, and the one bug that taught me to keep reading my diffs.
OpenAI Gave Away o3-mini, and I Think We Know Why
OpenAI made o3-mini free for everyone right as DeepSeek panic peaked, and the timing is not exactly subtle. Some notes after a weekend of poking at both.
Running DeepSeek R1 Locally Instead of Downloading the App
Skipping the DeepSeek app and the stock market panic entirely, I just ran R1's distilled models locally with Ollama on a three-year-old MacBook. Here's what that actually looked like.