On July 14, IBM’s stock opened down 24%, wiping out approximately $66 billion in market value in a single day. The direct cause was weaker-than-expected quarterly results. The CEO attributed the decline to enterprise clients reallocating IT budgets on a large scale towards AI infrastructure\u2014servers, storage, and memory\u2014while delaying software, consulting, and IT service contracts. The market is concerned that if overall corporate IT budgets do not grow correspondingly, traditional SaaS and software subscriptions will face sustained pressure, with simple, per-user priced products easily replaceable by AI agents being the first hit.<\/p>\n
Sources:<\/p>\n
Tencent Hunyuan released 1bit and 4bit quantized versions of its Hy3 model. The original 295B parameter model with BF16 weights was about 598GB. The 1bit version is compressed to 85.5GiB, fitting onto a single 96GB inference GPU. According to Tencent’s seven published benchmarks, the BF16 version averaged 71.0, the 4bit version about 69.5 (a 1.5-point drop), and the 1bit version about 68.1 (a 2.9-point drop), remaining usable for agent tasks, coding, and long-context understanding. With llama.cpp MTP speculative decoding, the 1bit version’s speed increased from 40.9 tok\/s to 61.7 tok\/s. This progress indicates flagship Mixture-of-Experts models are moving towards local deployment.<\/p>\n
Sources:<\/p>\n
OpenAI: Sam Altman announced GPT-5.6 Sol’s price was halved, with token efficiency for equivalent tasks being about double that of Fable. Agent products (Codex and ChatGPT Work) saw weekly usage grow 2.5x. Codex reached 7 million active users and received reset credits. Codex added interactive UI components, allowing visual tools to be embedded in conversations. GPT-5.6 Sol rose to 2nd place on Agent Arena.<\/p>\n
Anthropic: YC partner Tom Blomfield joined the Anthropic Compute team, with the lead stating “compute availability will be the most important issue.” Claude’s design lead was poached by Cursor. Anthropic launched a free premium version for US K-12 teachers. Analysis of 300,000 conversations revealed Claude exhibits different values across languages, with Chinese responses closest to “moderation”\u2014pointing out contradictions while maintaining a comforting tone. Multiple bloggers noted that without OpenAI providing a counterbalance to Anthropic, users would be worse off.<\/p>\n
Sources:<\/p>\n
Multiple bloggers discovered that the Grok Build CLI uploaded entire project codebases to xAI storage without clear user consent, potentially including secret keys. Sam Altman publicly commented it was “concerning.” Elon Musk responded that previously uploaded data would be completely deleted and reaffirmed that ZDR and \/privacy commands are effective. xAI issued a statement clarifying that user privacy is always respected. The same day, Musk shared news that Grok 4.5 ranked first on the Long-Horizon Terminal-Bench.<\/p>\n
Sources:<\/p>\n
OpenClaw released version v2026.7.1, featuring 3063 changes from 532 contributors. Updates include Web UI refactoring, iOS\/Android\/macOS app improvements, optimizations for Telegram\/Slack\/Discord channels, and support for models like GPT-5.6. Multiple bloggers released open-source tutorials for the WorkBuddy blueprint, covering installation to multi-agent workflows. Tokei v1.0.13 added WorkBuddy counting support. Some analysts noted that while OpenClaw is popular, its self-deployment complexity limits widespread adoption.<\/p>\n
Sources:<\/p>\n
Nearly 70-year-old Turing Award winner and reinforcement learning pioneer Richard Sutton announced the founding of Oak Lab with Khurram Javed. The goal is to develop AI agents capable of continuous learning from first-person experience, moving beyond static datasets to update knowledge through environmental interaction and trial-and-error. The company is advancing the OaK architecture, with a long-term vision of creating a trillion-parameter agent that requires only about 20 watts of power for real-time learning and planning.<\/p>\n
Sources:<\/p>\n
Sakana AI, in collaboration with the IT University of Copenhagen and Autodesk, developed 3D “cell bricks”\u2014intelligent modules that self-identify and self-repair through local communication alone, without a global controller. Dozens to hundreds of units can assemble into shapes like airplanes, guitars, or boats and infer the overall structure. If damaged, they can determine the repair direction. This is the first experiment to translate the natural principle of local rules generating global behavior from software simulation to real 3D hardware, remaining effective under signal interference and module failure.<\/p>\n
Sources:<\/p>\n
Stats: Timeline scans=360 Bloggers tracked=37 Total tweets tracked=212 Weighted tweet score=177.8 Original tweets=108 Retweets=29 Crawl attempts=2 Boundary coverage status=tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
IBM’s stock plunge reflects accelerated corporate IT budget shifts to AI infrastructure. Tencent’s Hunyuan model is compressed to 1bit for local deployment. OpenAI and Anthropic intensify competition on pricing, products, and talent.<\/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-1459","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\/1459","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=1459"}],"version-history":[{"count":0,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1459\/revisions"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1459"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1459"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1459"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}