World-record pivot.

Most use cases don't need fine-tuning. Before you look into fine-tuning, try the following: 1. Better prompts 2. Richer context 3. A more suitable model Fine-tuning is expensive, brittle, and often unnecessary. It should be the last thing you try, not the first.

Vibe-coding feels like magic. Until you're the one cleaning up the magic later.

You can probably solve 90% of the problems at a company with a combination of branching (if/else) and iteration (for/while). Stop overthinking and overengineering everything.

One of the amazing details of GPT-5 is the router that sits in front of the different models. It’s been a rocky start so far, but over time, this is one of the ideas that will last and become crucial going forward.

Today I woke up happy and more hopeful than the last 3 years. Humans are safe. My job is safe. If you build software, your future is bright.

Google, Anthropic, and xAI have an amazing opportunity to obliterate GPT-5 with their next release. I don’t think it will be that hard, honestly.

Well, something is clear: AGI is not happening any time soon. I can’t imagine how those people who quit learning feel right now…

This is cool and all but can we now go back to ChatGPT 4o while we wait for GPT 6?

Feels like a great day to repost this.