We recently started allowing candidates to use any AI tools during coding interviews. Today, we accidentally gave a data platform engineer a coding challenge meant for a full-stack engineer. It actually made the hire/no-hire signal much clearer.
I’ve always wondered why China open-sources so many LLMs. Then I realized they’re competing on energy. LLM inference turns electricity into intelligence, so the goal is to sell the cheapest intelligence via cloud inference. China is moving from manufacturing to intelligence.
Wish every Codex agent instance had a shared message index so each one knows what the others are changing in real time, flags overlapping work, and I don’t have to manually sync 2–3 parallel agents coding.
It probably makes sense for @OpenAI to put the browser directly into the chatGPT mac app. Search bar is the entry point for the browser UX for most people.
Most of my early-stage product code can be built with @OpenAI Codex within a few shots. Next I think Codex/CC should start to understand my business goal and proactively suggest improvements across the codebase to unlock new biz opportunities — aiming for staff engineer+ quality.
3 out of 5 of my @OpenAI Sora prompt failed due to IP protection. You lost most of the fun if you can’t remix household name IPs. The existing IP protection is close by default, but we need a paradigm shift and make it open by default for the AI era. The smartest influencers on Sora are already doing that by making their cameo public to anyone.