AI Industry Development & Policy Dynamics

According to The Block, Marathon Digital Holdings (MARA) agreed to acquire Long Ridge Energy & Power from FTAI Infrastructure for $1.5 billion on Thursday (April 29). The deal comprises the assumption of at least $785 million in existing debt, with the remaining balance paid in cash and backstopped

According to DeepSeek's technical report, on April 30, the company introduced Visual Primitives, a method that embeds basic visual units such as points and bounding boxes into reasoning chains to address the Reference Gap problem in multimodal tasks. The method reduces image token consumption

According to Beating, NVIDIA has released the weights for Cosmos-Reason2-32B, the flagship version of its physical AI reasoning vision-language model (VLM) designed to help robots and autonomous driving systems understand spatial, temporal, and physical principles. The 32-billion-parameter model,

According to reports monitored by Beating, Grok's web version is rolling out Imagine Agent Mode (beta), a new agent-based tool built on xAI's existing image and video generation capabilities. The interface replaces the traditional chat window with an infinite canvas and includes four preset

OpenAI’s official blog explains that Codex bans “banter goblins” and other creatures because the reward signal in nerd-persona training favored biological metaphors, leading to cross-persona contamination and RLHF misdirection. The incident was revealed by Barron Roth after a system prompt surfaced; OpenAI then used two strategies—short-term hard-coded fixes and long-term reward-signal removal—to warn about the fragility of reward design, and said post-training audits need to be more granular.

According to PANews, on April 30, Alibaba's Qwen announced the open-sourcing of Qwen-Scope, an interpretability module trained on Qwen3 and Qwen3.5 series models. The release covers 7 large language models across dense and mixture-of-experts variants, with 14 sets of sparse autoencoder

According to Reuters, Nvidia's B300 AI servers now sell for approximately 7 million yuan (US$1 million) in China, driven by a crackdown on smuggling networks and continued demand from local tech firms. The price has surged from about 4 million yuan (US$585,000) in late 2025, significantly

According to Chosun Daily, LG CNS reported on April 30 that first quarter operating profit increased 19.4% year on year to 94.2 billion won (US$64.1 million), driven by AI and cloud demand. Revenue rose 8.6% to 1.3 trillion won (US$894 million). The company's AI and cloud unit generated 765.4 billi

According to BlockBeats, on April 30, Billions Network has become the third-largest Agent ecosystem by deployment count, surpassing Ethereum with over 18,000 agents deployed. The network ranks behind only BNB Chain and Base. Billions has also synchronized its verified Agent identity feature across w

OpenAI announced the launch of GPT-5.5-Cyber, designed specifically for cybersecurity, and will deploy it in “the coming days” for key defensive units via a trusted access mechanism in collaboration with the government, not for open sale. In contrast to Anthropic’s Mythos open path, the White House opposes expanding it to 70 organizations. The Preparedness Framework rated it High (not yet critical), while strengthening cybersecurity protections. Going forward, attention will be on the deployment roster, whether CISA could be added, and the differences in security standards between the two sides.

According to Reuters, Schneider Electric reported first-quarter revenue of 9.8 billion euros (US$11.4 billion) on April 30, with organic growth lifted to 11.2% by AI data center demand. The French power and cooling equipment supplier said the result was slightly above consensus estimates and

According to TechCrunch Startup Spotlight, Protum, an AI governance startup, is raising a $2 million seed round aimed at closing by June 2026. Founded by Sandeep J., who brings 25 years of enterprise transformation experience, Protum provides a platform designed to give enterprises continuous

Research points to the externalities of AI layoffs: layoff costs are borne exclusively by companies, but the loss in consumer purchasing power is shouldered by the entire market. The more layoffs occur, the more demand shrinks—leaving both sides worse off. It proposes imposing an AI automation tax to internalize these external costs, and using tax revenues to fund retraining, restoring demand and stabilizing the economy.

Researcher Komatsuzaki said on X that mainstream LLM tokenizers impose a non-English language tax. In tests translating 《The Bitter Lesson》, Claude’s token increases in Hindi, Arabic, Russian, and Chinese were approximately 3.24×, 2.86×, 2.04×, and 1.71× respectively, clearly higher than OpenAI. China-based local models are more friendly to Chinese, indicating that training data skewed toward English creates cost inequality and has become a barrier to adoption.

On April 29, Microsoft reported third-quarter fiscal 2026 results for the period ended March 31, beating market expectations. Q3 revenue reached $82.886 billion, up 18% year-over-year and above the expected $81.4 billion; GAAP net profit grew 23% to $31.778 billion; non-GAAP diluted earnings per

Regulated B2B sportsbook supplier Kambi reported Q1 2026 EBITDA up 63.5% on the year. CEO Werner Bercher repeated his assertion that the entire 2026 FIFA World Cup will be priced and risk-managed by its AI trading system, with a new data point supporting his statement – Q1 bet automation is

OpenAI’s official announcement on April 29 said that its flagship developer conference, DevDay 2026, will be held on September 29 in San Francisco, returning to an in-person conference format after many years. It also announced a submission campaign: developers use GPT-5.5 and Image Gen to create works and submit them. Each week, Codex will select 2-3 creative submissions, and the submitters will receive free DevDay tickets (including cross-city airfare and hotel expenses). Conference theme: building a developer ecosystem around GPT-5.5 + Image Gen The core application stack for this DevDay is clearly centered on GPT-5.5. GPT-5.5 launched on April 23 and its API was fully opened on April 24; as of late April, GPT-5.4

Anthropic published a BioMysteryBench evaluation benchmark for AI bioinformatics analysis capabilities on April 29 in its official research announcement. It consists of open-ended questions from real research scenarios. The most notable data is: among the questions that a panel of human experts still could not solve, Anthropic’s flagship model Mythos solved 29.6%, while Opus solved 27.0% (4.7). Benchmark design: two tracks—questions solvable and unsolvable by experts BioMysteryBench is made up of two types of questions. The first is “solvable questions”—analysis tasks designed by bioinformatics researchers with standard answers for comparison. The second is “expert-unsolvable questions”—questions that human expert panels tried but still could not find a credible solution for, used to test whether models can go beyond the boundaries of current domain knowledge.

According to Cointelegraph, Sam Altman announced OpenAI will launch GPT-5.5-Cyber, a cybersecurity model, in the coming days. The model will be made available to critical cybersecurity defenders, with OpenAI planning to establish trusted access mechanisms in collaboration with government entities to

According to Sequoia Capital's video series featuring Demis Hassabis, Google DeepMind's CEO predicts artificial general intelligence (AGI) will arrive by 2030. Hassabis, a 2024 Nobel Prize winner in chemistry, views this timeline as a strict business reality that forces leaders to rethink product de