Keep up with tech in 5 minutes
Get the free daily email with summaries of the most interesting stories in startups 🚀, tech 📱, and programming 💻!
Join 1,600,000 readers for one daily email
Unlocking asynchronicity in continuous batching (20 minute read)
Asynchronous batching can reduce idle time between CPU and GPU cycles, improving GPU utilization for inference by 22%. By using CUDA streams and events, CPU tasks prepare batch N+1 during batch N's GPU computation, eliminating idle gaps. This method yields more efficient GPU operations without changing kernels or models, enhancing generation speed substantially.
Apr 29 | Blog
The Trust Problem With AI Agents
Why developers should not entirely rely on agents, and what you can do about it.
Sponsored
May 15 | Tech
The great memory panic of 2026 (6 minute read)
Memory prices could move from 15% to 40% of the bill of materials for Apple devices. Apple operates at an enormous scale, and each of its products consumes some memory. When dealing with this kind of scale, there's usually a couple of years of lead time. The spike in demand for memory has disrupted this pipeline. Apple still has a lot of room to negotiate price, and the whole situation could end up with Apple gaining a lot of market share.
May 15 | Product
IC work is the new career flex (7 minute read)
AI is making it possible for senior individual contributors to drive major business impact without managing a team. The best “High-Impact ICs” use AI to build, test, and ship work end to end, making management a choice rather than the only path to influence.
May 15 | DevOps
Supertonic (GitHub Repo)
Supertonic 3 is a new open-source text-to-speech system with 99M parameters that runs entirely on-device without cloud APIs and supports 31 languages while delivering competitive accuracy against models up to 20x larger. The lightweight system runs fast on CPU, works offline on devices from Raspberry Pi to e-readers, and is available now through Python, Node.js, browser implementations, and mobile platforms.
