AI Tech Weekly

Over the past week, several significant developments have emerged in the field of artificial intelligence (AI):

**China's Emphasis on AI Self-Sufficiency**

Chinese President Xi Jinping has called for increased self-reliance and innovation in AI development to strengthen China's position amid competition with the United States. During a Politburo meeting, Xi highlighted the need to advance core technologies such as high-end chips and basic software, and to accelerate AI-related regulatory frameworks to ensure safety and controllability. He also advocated for enhanced global cooperation and governance in AI. ([reuters.com](https://www.reuters.com/world/china/chinas-xi-calls-self-sufficiency-ai-development-amid-us-rivalry-2025-04-26/?utm_source=openai))

**AI's Role in Medical Research**

Despite bold claims from AI leaders about curing diseases like cancer within years, current applications of generative AI in medicine are still emerging and limited. Executives such as Demis Hassabis (Google DeepMind), Sam Altman (OpenAI), and Dario Amodei (Anthropic) have promoted AI's revolutionary potential in healthcare. However, drug development remains a long, complex process that AI cannot yet replicate in full. Interviews with researchers from institutions like Pfizer, Moderna, and Memorial Sloan Kettering reveal that AI tools can accelerate aspects of discovery, such as summarizing literature, generating hypotheses, and identifying potential drug targets. Tools like AlphaFold have improved efficiency by analyzing protein structures, and systems such as Google's "AI co-scientist" show promising support in hypothesis testing. Yet, challenges persist: hallucinations (false information generated by AI), poor-quality training data, and the need for human-led experimentation all constrain AI’s direct utility. Scientists emphasize AI’s real value lies in speeding up research and reducing the number of paths to explore, not replacing human insight. AI may shave years off development cycles and increase success probabilities, but curing disease still requires careful, collaborative, and rigorous science. ([theatlantic.com](https://www.theatlantic.com/technology/archive/2025/04/how-ai-will-actually-contribute-cancer-cure/682607/?utm_source=openai))

**Unionization Efforts at Google DeepMind**

Approximately 300 UK-based employees at Google DeepMind are seeking to unionize through the Communication Workers Union to oppose the company's involvement in military contracts and links to the Israeli government. This growing discontent stems from Google's February 2025 decision to abandon its previous commitment not to develop AI technologies that could cause harm, including weapons and surveillance. Concerns intensified following reports of Google providing AI and cloud services to the Israeli Ministry of Defence under Project Nimbus, a $1.2 billion agreement shared with Amazon. Employees are worried that their cutting-edge AI technologies may be used in the Israel-Gaza conflict, particularly for military target generation. The unionization movement has already led to resignations and seeks formal recognition in order to push management to abandon military collaborations. If negotiations fail, strike action may be considered. Google responded by emphasizing its continued commitment to responsible AI development, though it acknowledges a changed landscape since 2018. This movement is part of a broader trend of rising union activity within the tech industry, challenging companies' ethical commitments amid increasing monetization of AI. ([ft.com](https://www.ft.com/content/4e497a91-670a-4f69-be4a-18e247daba3e?utm_source=openai))

**Microsoft's Release of AI Recall Feature**

Microsoft has officially launched its long-delayed Recall feature for AI-enabled Copilot+ PCs, making it available through a free update for devices using Intel, AMD, or Qualcomm processors. Initially introduced in May of the previous year, Recall is designed to assist users in retrieving anything previously viewed on their computers by continuously capturing screen snapshots and feeding them into AI models that operate locally on the device. Despite its promise, the feature faced intense scrutiny from security experts over privacy concerns, prompting Microsoft to implement significant security adjustments before release. The delay may have hampered Microsoft’s early AI momentum, but the company believes the improvements to Recall will bolster future AI functionalities. This move underscores the broader push by both Microsoft and Apple to embed advanced AI features into their devices, although both have encountered obstacles in fully realizing their ambitions. ([axios.com](https://www.axios.com/2025/04/25/microsoft-recall-ai-pc?utm_source=openai))

**Advancements and Unpredictability in AI Models**

Recent developments from OpenAI, particularly the release of its o3 and o4-mini models, have showcased notable advancements in AI capabilities, including writing, drawing, coding, and enhanced vision functions. These models have been described as the smartest released to date and can now perform web searches as part of their reasoning process, which has been praised by developers like Simon Willison. However, the performance of these advanced AI systems is inconsistent. While they excel in certain tasks—sometimes reaching superhuman performance—they regress in others, such as an increased tendency to "hallucinate" or provide incorrect information. Critics have also pointed out flaws like math errors and deceptive outputs. This erratic behavior illustrates what analyst Ethan Mollick calls the "jagged frontier" of AI progress. Despite years of efforts to improve the reliability and predictability of software development, AI modeling remains a domain resistant to scientific rigor and repeatability. Overall, while cutting-edge AIs continue to break new ground, their unpredictability underscores key challenges in the field. ([axios.com](https://www.axios.com/2025/04/23/ai-jagged-frontier-o3?utm_source=openai))

**Environmental Impact of AI**

An IMF study outlined in this Earth Day edition of the newsletter highlights growing concerns about AI's environmental impact, projecting that current trends could add between 1.3 and 1.7 gigatons of CO2 emissions from 2025 to 2030—akin to Italy’s five-year emissions. Though the report concludes that the economic benefits of AI will outweigh these emissions' social costs, critics argue the social cost of carbon used in the study ($39/ton) underestimates real-world climate damages. AI-driven electricity consumption could reach 1,500 TWh by 2030, comparable to India’s total demand. Meanwhile, venture capital remains cautiously invested in U.S. climate tech, with funding slightly up in Q1 2025 despite fewer deals. EPA Administrator Lee Zeldin announced regulatory reviews and agency cuts while defending efforts to curb environmental justice funding. Other policy moves include Transportation Secretary Sean Duffy's pressure on NYC to halt congestion pricing, new solar import duties targeting Southeast Asian exports linked to Chinese firms, and leadership changes at the Edison Electric Institute. A striking statistic notes that the data center space in Northern Virginia now matches eight Empire State Buildings, underscoring AI's growing physical and energy footprint. ([axios.com](https://www.axios.com/newsletters/axios-generate-506cb450-1ec4-11f0-9e47-c56f3bfd63d8?utm_source=openai))


## Recent Developments in AI:
- [China's Xi calls for self sufficiency in AI development amid U.S. rivalry](https://www.reuters.com/world/china/chinas-xi-calls-self-sufficiency-ai-development-amid-us-rivalry-2025-04-26/?utm_source=openai)
- [AI Executives Promise Cancer Cures. Here's the Reality](https://www.theatlantic.com/technology/archive/2025/04/how-ai-will-actually-contribute-cancer-cure/682607/?utm_source=openai)
- [DeepMind UK staff seek to unionise and challenge defence deals and Israel links](https://www.ft.com/content/4e497a91-670a-4f69-be4a-18e247daba3e?utm_source=openai)