{"id":165351,"date":"2026-03-19T23:12:07","date_gmt":"2026-03-19T21:12:07","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/vvedyon-v-stroj-ai-servis-sashiko-dlya-reczenzirovaniya-izmenenij-v-yadre-linux"},"modified":"2026-03-19T23:12:07","modified_gmt":"2026-03-19T21:12:07","slug":"vvedyon-v-stroj-ai-servis-sashiko-dlya-reczenzirovaniya-izmenenij-v-yadre-linux","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/vvedyon-v-stroj-ai-servis-sashiko-dlya-reczenzirovaniya-izmenenij-v-yadre-linux","title":{"rendered":"The AI service Sashiko for reviewing changes in the Linux kernel has been put into operation","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Roman Gushchin from the Linux kernel development team at Google has announced the creation of a new code review system utilizing large language models. Development has been ongoing for the past few months, and it has been named Sashiko, after a traditional Japanese stitching technique consisting of small straight stitches that create various patterns. <\/p>\n
The project has been in use at Google for some time to identify issues and is now available to everyone, configured to automatically review all patches submitted to the Linux kernel developers' mailing list. The Sashiko code is written in Rust and is open under the Apache 2.0 license. The system is self-sufficient and can be used on personal hardware. <\/p>\n
Sashiko was designed to work with the Google Gemini Pro 3.1 model but has been partially tested with Claude and is likely to work with other modern large language models. The prompts used for reviewing are based on a set of review-prompts prepared by Chris Mason, the creator of the Btrfs file system. The token budget and infrastructure for Sashiko are funded by Google. The project rights have been transferred to the Linux Foundation. <\/p>\n
According to the tests conducted, using the Gemini 3.1 Pro model, the Sashiko toolkit was able to detect 53% of errors from an unfiltered set, which included 1000 recent issues in the core marked with the tags «Fixes:». At first glance, 53% may not seem particularly impressive, but it should be noted that all of the identified issues were initially overlooked during human review. \n Source: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"