Xiaoliu BOT
中文EN

X Platform June 12 AI Brief | SpaceX Record IPO, Multiple AI Models Released Densely, OpenAI Acquisition Strengthens Execution Capabilities

SpaceX completes the largest IPO in history, global AI models compete to release new versions, and the Chinese tech community discusses the technical roadmap and commercialization challenges of Huawei’s Pangu large model.

SpaceX Completes Largest IPO in History at $75 Billion

SpaceX (ticker SPCX) officially listed on Nasdaq yesterday, raising $75 billion at $135 per share, making it the largest IPO ever. Goldman Sachs and Morgan Stanley served as lead underwriters. Elon Musk thanked the underwriting team in a post and forwarded confirmation messages from multiple institutions. This IPO will create approximately 4,400 new millionaires, spanning roles from engineers to cafeteria staff. SpaceX also advanced Starlink launch missions, with bases in Florida and California launching 29 and 24 Starlink satellites respectively.

Sources:

Multiple AI Companies Release New Models or Technical Progress

MiniMax-M3 is officially open-sourced, a native multimodal model with 428 billion parameters supporting a 1 million context length. The model undergoes mixed-modal training from the first step and uses sparse attention mechanisms to improve long-context efficiency. GLM5.2 is expected to be released this month, supporting 1 million context and multimodality. Tencent AI also plans to release more than one new model this month. DeepSeek V4.1 and Kimi K2.7 (hinted) may launch soon.

Sources:

Huawei Pangu Revealed as China’s First True LLM Architecture, but Commercialization Path Questioned

Blogger @MaxForAI dug into a 2022 Huawei Cloud article, revealing that the team led by Huawei Cloud AI Chief Scientist Tian Qi started the Pangu large model project in September 2020, officially releasing the industry’s first 110 billion parameter Chinese pre-trained model in April 2021, and upgrading to 200 billion parameters in May of the same year. The Pangu NLP large model uses 40TB of text data, and the Pangu CV large model implements on-demand extraction and contrastive learning techniques. However, the first large model to complete filing in China was Baidu’s ERNIE, raising questions about the technical roadmap or commercialization strategy after Pangu 3.0.

Sources:

OpenAI Acquires Ona to Complete Codex Execution Layer Capabilities

OpenAI announced the acquisition of cloud development environment company Ona (formerly Gitpod), which will be integrated into the Codex team after the transaction closes. The acquisition aims to provide Codex with a secure, isolated, long-term online cloud execution environment, allowing AI agents to continue running tasks after laptops are closed, handling codebases, CI, internal tools, and other enterprise-level work. This indicates that the competition for programming assistants is shifting from code generation capabilities to the ability to securely execute tasks on behalf of organizations.

Sources:

Chinese Tech Bloggers Discuss Big Company Layoffs and Economic Downturn Coping Strategies

Blogger @imwsl90 mentioned Alibaba “sending high-end talent to society,” quoting Jack Ma’s remark that “earning 30,000-50,000 a month and having a job is a blessing,” lamenting the current difficult economic environment. The same blogger advised independent entrepreneurs to first learn marketing and sales skills, criticizing self-media that encourages lying flat. @MaxForAI mocked the situation where big companies lay off people and then “beg front-end brothers to come back,” reflecting changes in supply and demand for tech positions. Multiple bloggers discussed that during economic downturns, one should focus on building income systems rather than lying flat.

Sources:

AI-Assisted Development Practice Shows 1.5 Hours from Requirements to Submission

Blogger @imwsl90 shared an experience using Claude to develop a Chrome extension “Tab Limiter”: starting AI code writing at 1:00 PM, submitting for review at 2:23 PM, completing the entire process from requirements to submission in just 1.5 hours. The extension features include limiting the number of tabs and automatically closing tabs when exceeding the limit, with code, logo, and instructions all generated by AI. This demonstrates the efficiency of current AI in simple tool development, though the functionality is relatively basic.

Sources:

Tesla FSD 14.3.4 Begins Rolling Out, Belgium Becomes 5th European Approval

Elon Musk confirmed that Tesla FSD 14.3.4 version has started rolling out, with Belgium becoming the fifth European country to approve FSD Supervised, following the Netherlands, Lithuania, Estonia, and Denmark. Tesla’s Optimus robot is serving at a Tesla restaurant, showcasing the application of robots in dining scenarios. Musk also forwarded discussions about Grok’s authenticity and the working mechanism of X Community Notes.

Sources:

Statistics: Scan timeline lines=360 Hit bloggers=48 Total hit tweets=232 Weighted tweet score=180.85 Original tweets=116 RT tweets=55 Fetch attempts=2 Boundary coverage status=tail_confidently_crossed_target_boundary