SpaceX completed the 12th test flight of Starship. The V3 heat shield withstood peak temperatures of approximately 1450\u00b0C (2600\u00b0F) during re-entry, with Elon Musk confirming “heat shield performance is good.” The Starlink satellite network enabled the first-ever full 4K live stream from launch to splashdown for this test flight. Multiple photographers deployed 7 cameras near the launch pad, all capturing clear photos, with @AJamesMcCarthy’s work being widely shared. The Super Heavy booster’s thrust is equivalent to about 80 Boeing 747s operating simultaneously. Musk also revealed that Boring Company’s Prufrock tunnel boring machine has achieved continuous boring and pipe-pushing synchronization, with robots from Las Vegas to Dubai being remotely controlled from the Austin operations center.<\/p>\n
Sources:<\/p>\n
OpenAI Codex recently experienced issues with quotas depleting too quickly. Official account @thsottiaux confirmed the cause was a previously rolled-back optimization affecting cache hit rates for long sessions. The issue has been fixed and usage quotas have been reset for all accounts. Several Chinese bloggers expressed pleasant surprise and shared numerous Codex usage tips: @steipete demonstrated running Codex autoreview for 5 hours to automatically fix issues during a major refactor and built a Cloudflare-based cloud Codex environment; @LufzzLiz optimized a set of Codex self-evolving prompts to allow Agents to automatically inventory repetitive workflows and create skills; @vista8 shared complete prompts for optimizing Mac network speed using Codex.<\/p>\n
Sources:<\/p>\n
The official vLLM project account disclosed that a trainee, guided by a training institution, recently submitted a PR addressing a non-existent issue to the project and spammed @mentions to maintainers to gain attention. The contributor has been banned. @MaxForAI provided an in-depth analysis: training institutions want case studies, trainees want resume material, maintainers are forced to spend time screening, and the open-source community bears the cost of spam PRs. He warned that if this behavior spreads, large projects will become more defensive, new contributor PRs will be harder to trust, and those genuinely wanting to participate in open-source will be negatively affected.<\/p>\n
Sources:<\/p>\n
@imwsl90 shared news of OpenCode reaching $10 million in Annual Recurring Revenue (ARR). He believes OpenCode, like OpenRouter, is essentially in the business of selling tokens, and the model of using open-source for traffic and subscriptions for profit has been validated. Users tend to directly use OpenCode’s built-in plans rather than subscribing to large models themselves and configuring them. Pieter Levels’ earnings on X this month exceeded $14,000 for the first time, surpassing his earnings from a single product, giving him his first experience of earning purely from posting on X as a major influencer.<\/p>\n
Sources:<\/p>\n
@oran_ge referenced Baoyu’s translated article “DeepSeek’s $10 Trillion Grand Strategy,” pointing out that while DeepSeek V4 Pro may not be the best model, its cache technology is essentially free, a technology needed by all large models. Even Opus could see costs drop tenfold using this tech. He also believes V4.1 will improve rapidly after being trained with real harness data. iFlytek’s 39-yuan tier has also made GLM 5.1 available.<\/p>\n
Sources:<\/p>\n
@MANISH1027512 trained a style using 200+ images, declaring “MJ v8.1 is so back.” He admitted being very disappointed when v8 was first released, even canceling his subscription, but was persuaded by VSC community members not to give up, ultimately finding v8.1 showed clear improvement. Several users agreed. Meanwhile, @cellinlab shared that GPT Image 2’s cover image creation capability has greatly increased, leading to a phenomenon where some bloggers’ “cover images and video content are mismatched.”<\/p>\n
Sources:<\/p>\n
@MaxForAI discovered Claude finally allows appeals, speculating Anthropic might be buckling under pressure, but stated “already in love with Codex.” @levelsio complained that Anthropic daily forces his usage tier back to medium, calling it annoying. Gorden Sun summarized subscription usage across platforms in third-party Agents: ChatGPT subscriptions work with multiple Agents, Grok also works on multiple platforms, but Claude subscriptions cannot be used with any third-party Agents, and its own Agent SDK changed from shared subscriptions to dedicated quotas.<\/p>\n
Sources:<\/p>\n
@LufzzLiz released a complete practical tutorial on “Building a Trustworthy Architecture Wiki for Any Open-Source Project,” including skill templates, practical cases, and principle explanations. He suggested combining it with Agents’ scheduled task capabilities for continuous wiki iteration. Additionally, @vista8 and Teacher Yao announced a GEO public course at waytoagi at the end of the month, sharing systematic logic and core strategies for search engine optimization.<\/p>\n
Sources:<\/p>\n
OpenClaw released version 2026.5.22, with main improvements including: more streamlined Gateway\/model startup paths, \/models command reduced to ~5ms, npm package dependency locking, and reinforced Windows installation and update paths. The chat session selector in Control UI added search and pagination features. \/verbose on can now be used in private chats, groups, channels, and forum topics.<\/p>\n
Sources:<\/p>\n
Stats: Timeline scans=240 Bloggers matched=24 Total tweets matched=121 Weighted tweet score=98.2 Original tweets=59 RT tweets=22 Crawl attempts=1 Boundary coverage status=tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
SpaceX successfully completes the Starship V3 test flight with full 4K live stream. OpenAI fixes Codex issues and resets quotas, sparking practical tips discussions. The vLLM project exposes fake PR submissions guided by training\u2026<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[19],"class_list":["post-1290","post","type-post","status-publish","format-standard","hentry","category-brief","tag-x--ai-"],"_links":{"self":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1290","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/comments?post=1290"}],"version-history":[{"count":0,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1290\/revisions"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1290"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1290"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1290"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}