Lumis Daily Briefing — Sep 19, 2026 — OpenAI used its own LLMs to design a chip, and the AOSP walls are closing in
OpenAI used its own LLMs to help design its Jalapeño AI chip
AI designing AI hardware closes the loop on self-improving infrastructure. This could compress chip design cycles from years to months industry-wide.
Researchers show LLMs can talk directly through cached hidden states, not text
Bypassing tokenized text for KV-cache communication could slash multi-agent LLM latency and cost. A step toward native machine-to-machine AI protocols.
Android 17 breaks tradition, adds new APIs without open-sourcing them to AOSP first
GrapheneOS warns this sets precedent for Google withholding core platform code, threatening custom ROMs and open-source Android's future viability.
Cloudflare cuts another 100TB of RAM usage with clever math, not just hardware
Algorithmic memory optimization at Cloudflare's scale translates directly into massive infrastructure cost savings, a model for other hyperscalers.
Ledger researchers crack RP2350 secure debug using photon-emission laser attacks
Physical fault injection defeating a chip's secure boot shows hardware security still lags software. Relevant to anyone shipping secure IoT or crypto hardware.
Stanford study finds the human brain functions as two separate organs
Rethinking basic brain structure could reshape neuroscience-driven approaches to brain-computer interfaces and neuromorphic AI design.
Cloudflare Quick Tunnels lets anyone expose a local server instantly, no signup
Frictionless tunneling tool lowers the barrier for developers to demo and share local apps publicly, competing directly with ngrok.
Get tomorrow's briefing delivered at 07:00 UTC.
Lumis synthesizes the top AI and tech developments into a sharp 3-item briefing — personalized to your sources, delivered before your day starts.
7 days free. Cancel anytime. No credit card needed.