1. Cognition AI Raises Over $1 Billion, Valued at $26 Billion
According to Bloomberg, Cognition AI — the company behind Devin, the “world’s first AI software engineer” — has completed a new funding round of over $1 billion, reaching a post-money valuation of $26 billion, making it the most highly valued AI coding company globally. The round was co-led by Lux Capital, General Catalyst, and 8VC, with participation from Ribbit Capital, Atreides Management, and Founders Fund. Just eight months after its previous round at $10.2 billion, Cognition’s valuation has grown 2.5×. The three Chinese founders collectively hold five International Olympiad in Informatics gold medals and built Devin’s prototype from a short-term rental apartment.
AI Pulse View: A $26 billion valuation signals that the market is positioning AI coding assistants as the future infrastructure of software engineering, not just efficiency tools. The co-leadership of Lux Capital (hard tech infrastructure), General Catalyst (enterprise transformation), and 8VC (government-scale deployment) represents three converging narratives —预示着 AI Agent is evolving from developer tool to enterprise platform.
Source: 36Kr, June 1, 2026
2. Alibaba “Shelves Products” vs ByteDance “Trains Skills”: China’s Tech Giants Diverge on AI Strategy
Two closely timed events in late May revealed fundamentally different AI approaches from China’s tech giants. Alibaba has fully integrated its Qianwen app with Taobao, accessing 4 billion product listings and 20 years of shopping data. Ant Group’s AI payment product became the world’s first to surpass 100 million users and transactions. Meanwhile, ByteDance’s Jimeng AI Seedance 2.0 topped video model benchmarks, and the company’s 2026 capital expenditure ceiling reaches 470 billion yuan (~$70 billion), potentially scaling to $100 billion in ideal conditions.
AI Pulse View: Alibaba is building AI-era “utilities and retail checkout,” emphasizing rapid integration and commercialization. ByteDance is building an AI-era “Nobel lab,” pursuing foundational research breakthroughs and long-term infrastructure investment. These approaches represent the classic divide between “application-driven” and “technology-driven” paths in AI industrialization.
Source: 36Kr, May 31, 2026
3. Zig Programming Language Says “No” to AI Code: Creator Calls AI Contributions “Garbage”
The open-source modern programming language Zig explicitly bans AI-assisted code contributions. Creator and lead developer Andrew Kelley called AI-assisted contributions “garbage” on a JetBrains podcast, noting they consume limited code review time and slow down the entire team. Kelley emphasized that mentorship is a core mission of the Zig project, and AI contributions work against this goal. Ironically, Bun — a project built in Zig and acquired by Anthropic — embraces AI, with its creator recently using Claude Code to port Bun from Zig to Rust.
AI Pulse View: Zig’s stance raises deeper questions about the balance between AI code generation quality and developer growth. As tools like Claude Code and OpenAI Codex sweep Silicon Valley, Zig represents an alternative voice: code review isn’t just quality assurance — it’s人才培养. The implicit cost of widespread AI adoption may be the erosion of developer community health and the learning process.
Source: 36Kr/Machine Heart, May 31, 2026
4. Unitree Robotics IPO: Robots Emerges as Semiconductor’s Next Super Terminal
On June 1, 2026, Unitree Technology is heading to the STAR Market for its IPO listing review. Beyond being a robotics company’s capital market milestone, this signals a paradigm shift for the semiconductor industry: embodied AI is no longer satisfied with selecting from existing chip catalogs. Instead, it’s demanding custom designs around motion control, millisecond-level real-time response, extreme low power consumption, and multi-sensor fusion — pushing chips and electronic components to be redesigned. Unitree splits embodied AI into “brain” (perception and decision) and “cerebellum” (motion control), representing two entirely different chip technology paths.
AI Pulse View: Robots becoming semiconductors’ “super terminal” is forward-looking. Following smartphones, cars, and servers, embodied AI could become the fourth major driver of chip innovation. The算力 race in the “brain”赛道 (Horizon, Cambricon, Black Sesame) and the precision control布局 in the “cerebellum”賽道 (Renesas, Rockchip) preview an entirely new semiconductor growth curve.
Source: 36Kr, May 31, 2026
5. Anthropic Bans AI Tools During Job Interviews to Assess “Real Thinking”
According to Bloomberg Businessweek, Anthropic explicitly bans candidates from using AI tools during live interviews unless otherwise stated. Anthropic’s hiring process includes up to five rounds, featuring a rigorous “culture interview” where candidates face questions about values, worldview, and ethical dilemmas. Failing this round essentially eliminates a candidate’s chances. Both Anthropic and OpenAI are currently minting dozens of multimillionaires, with top salaries reaching $850,000 plus equity.
AI Pulse View: The policy reveals an interesting paradox: a company building the most advanced AI bans its use during hiring. This signals that independent thinking ability and cultural fit remain irreplaceable by AI when evaluating core talent. As AI tools become ubiquitous in software development, distinguishing “ability to use AI” from “independent thinking ability” will become a new challenge in recruitment.
Source: The Decoder, May 31, 2026
6. Study: AI Search Agents Confirm What They Already Know Instead of Actually Researching
Researchers from Harbin Institute of Technology and Xiaohongshu found that frontier models like GPT-5.4, Gemini 3.1 Pro, and Claude Sonnet 4.6 achieve high scores on BrowseComp not because of genuine search ability, but due to “intrinsic knowledge dependence” (IKD) — relying on internal knowledge absorbed during training. When researchers removed supporting documents from the search index, all models performed worse than without any tool access at all — MiniMax M2.5 dropped from 44.5% to 8.0%.
AI Pulse View: This research reveals a fundamental flaw in AI search agents: searching often confirms existing hypotheses rather than exploring new knowledge. When search results conflict with the model’s internal knowledge, it tends to ignore the evidence. This raises serious questions about the reliability of AI search products and suggests the need for better “knowledge frontier” benchmarks to truly measure search capability.
Source: The Decoder, May 31, 2026
7. Anthropic Study: Men Use AI Coding Agents More Than Twice as Often as Women in Social Science
An Anthropic study found that in social science research, men use AI coding agents more than twice as frequently as women. This gap reveals a gender divide in AI tool adoption that could have profound implications for the future distribution of AI skills and career opportunities.
AI Pulse View: If the gender gap in AI tool usage continues to widen, it could exacerbate existing inequalities in the tech sector. Early adopters of AI coding tools will gain significant advantages in efficiency and capability, requiring educational institutions and enterprises to focus on inclusivity and equity in AI tool promotion.
Source: The Decoder, May 31, 2026
8. Large-Scale Study: Making AI Chatbots More Helpful Weakens Their Ability to Simulate Human Behavior
A large-scale study found that optimizing AI chatbots for “helpfulness” significantly reduces their ability to simulate human behavior. When models are trained to be more helpful and agreeable, their capacity to generate diverse, human-like responses diminishes.
AI Pulse View: This finding reveals a core tension in AI alignment research: the trade-off between usefulness and authenticity. If the goal is to have AI more accurately reflect human behavior in simulated conversations, psychological research, or social experiments, over-optimizing for “helpfulness” may be counterproductive. This has important implications for AI applications in social science and human-computer interaction research.
Source: The Decoder, May 30, 2026
Other Updates
- OpenAI Codex now supports autonomous Windows PC operation for bug hunting and app testing (The Decoder, May 31)
- Microsoft & Nvidia reportedly teaming up on AI PCs that run actual agents instead of Copilot assistants (The Decoder, May 31)