Anthropic Surges Past OpenAI to Become the World's Most Valuable AI Startup

Anthropic has reportedly overtaken OpenAI in valuation, becoming the world's most valuable AI startup. The story has drawn significant attention, picking up 138 points and 117 comments on Hacker News.

OpenAI has long been the iconic player of the AI industry, so this appears to reflect recognition of Anthropic's safety-first stance and the steady expansion of enterprise demand. That said, direct comparisons of the two companies' business models and revenue scale call for caution — valuation cannot simply be equated with capability. How far market expectations are backed by actual results remains to be seen as things unfold.

An Experiment Where AI Models Ran Virtual Societies: Claude Built the Safest One, Grok's Went Extinct

Researchers had multiple LLMs run simulated societies, and the reported results show that Claude built the safest society, while the society run by Grok (made by xAI) ended in extinction.

The experiment is an attempt to visualize AI models' "values" and "decision-making tendencies" at the scale of a society. The results call for careful interpretation, but it is intriguing that they suggest each model's alignment policy manifests as observable differences even in simulations of collective behavior.

AI Chip Demand Is Straining Wafer Capacity, Dealing a Blow to the Budget PC Market

According to research by TrendForce, demand for AI semiconductors is absorbing wafer production capacity, and as a result, the budget segment of the PC market is taking a heavy hit.

The picture is one of AI growth squeezing the existing semiconductor supply-demand balance, with the impact rippling all the way down to low-cost consumer devices. Some also point out that the gap in hardware access could widen between those who benefit from AI and those who do not.

Meta's AI Business: Struggling to Monetize Beyond Ads?

CNBC asks, "Meta has struggled with non-ad sales — can AI change that?" The topic is also sparking discussion on Hacker News, with some arguing that Meta's ad-dependent nature is casting a shadow over the monetization of its AI products as well.

While Meta wields considerable influence in the AI world through the open-source release of its Llama series, how to monetize its own AI products remains an open challenge.

QEMU Moves to Reconsider Its Ban on AI-Generated Code Contributions

The Register reports that QEMU, the open-source emulator project, has begun considering a review of its policy, which has until now banned contributions of AI-generated code.

How open-source communities handle AI code varies widely from project to project, with approaches ranging from outright bans to conditional acceptance. QEMU's change of direction appears to reflect an on-the-ground shift in perception that the quality of AI-generated code is steadily improving.

The History of LLM Prompting Has Changed — Before and After Reasoning Models

A noteworthy article was published in Japan's tech community. Titled "LLM プロンプトと変数の扱い方が変わった — 4 年間の歴史振り返り" (How LLM Prompts and Variable Handling Have Changed — A Four-Year Retrospective), it traces the story from Chain of Thought in 2022 through how the arrival of reasoning models fundamentally transformed what prompting should look like.

The observation that 2022–2024-era templates like "think step by step" and "lower the temperature for reproducibility" no longer work with today's reasoning models (GPT-5 Thinking, Claude's extended thinking, the DeepSeek R series, and so on) carries an important lesson for practitioners.

---