If GPT 5.6 beats Opus 4.8 on coding, Anthropic will have no choice but to put Fable 5 into its subscription tier.
Otherwise, developers will move to Codex overnight.
After 18 months of writing, coding, and experimenting, Build a Reasoning Model (From Scratch) is
finally out!
My first copies just arrived! 📚
440 full-color pages. Inference scaling, reinforcement learning, and distillation from scratch.
Arguing with AI is surprisingly useful.
Sometimes I start with an idea, and it pushes back hard enough that I have to admit I missed something.
Then after thinking a bit more, I find a gap in its reasoning and push back again.
That back-and-forth is the fun part.
AI is not always right. But it is a very good thinking sparring partner.
Sometimes I miss the very early ChatGPT web app.
It was raw, limited, and often unstable. People were trying all kinds of prompts, pushing the boundaries, and half-joking about getting banned.
Looking back, that period felt special not because the product was better, but because it was not fully defined yet.
Users, rules, use cases, and expectations were all being formed in real time.
Mature products are more reliable, of course. But they also lose some of that early wildness.