GoKawiil
Tech News
DeepSeek has shipped V4.1 Flash 1017, dated October 17, 2026, as a snapshot of its V4.1 Flash model available via the deepseek-flash API. The sparse mixture-of-experts model uses a new Causal Encoder-Decoder architecture, activating 8B parameters on input and 16B on output from a 552B-parameter backbone, with native image understanding trained jointly with text from the start.
Neuralwatt CEO Chad Gibson analyzed a month of cloud traffic and found that different agent harnesses produce wildly different prompt-cache hit rates, from 1.6% to 11.1% uncached input tokens. The study found that a cache miss roughly doubles a request's energy use, and that deep sessions past 128k tokens with major prompt re-reads can cost as much compute as roughly 600 normal turns at that depth.
A technology explainer details how safely removing external drives depends on the operating system, noting that Windows 10 defaults removable drives to a 'Quick removal' policy that reduces risk but lowers performance. It warns that Mac background tasks like Time Machine and Spotlight, and Linux's reliance on write caching, can make sudden disconnection more likely to cause data loss or corruption.
Cloudflare has launched support for the HTTP Vary response header within its Cache Rules feature, available to customers on every service tier. The update lets origin servers specify which request headers may affect a response, while giving Cloudflare customers control over how those variations are handled in caching—whether by normalizing header values, preserving exact values, or bypassing the cache entirely.
A Show HN submission presents an interactive atlas of system designs, breaking down technologies like PostgreSQL, Redis, and Memcached into their core features and the specific scenarios where each is the right choice. Each entry includes architecture diagrams and defines related concepts such as consistency, caching, and rate limiting. The project appears aimed at helping engineers quickly identify which data store fits a given technical requirement.
The team behind the open-source Bomfather eBPF security agent found that the costliest part of their file-access enforcement wasn't applying allow/deny decisions, but repeatedly walking parent directories to determine which path-based policy applied to a given file. By caching the resolved policy per inode instead of recalculating it on every file open, they cut kernel-side CPU overhead by roughly 90%, particularly benefiting workloads like databases that repeatedly access the same file paths.
A researcher replayed 68,266 requests from 393 Claude Code sessions and 23,608 Mooncake requests through a custom prefix-cache simulator, testing three approaches meant to outperform the standard LRU eviction policy used by vLLM, SGLang, LMCache and Mooncake Store. None of the alternatives beat the default LRU-over-radix-leaves baseline, and the analysis found that cache misses mostly come from rapid, seconds-apart tool-calling loops rather than sessions timing out after going idle.