Sunday, May 31, 2026

What happened: Anthropic conducted two major findings on AI adoption and hiring: the company bans AI tools during job interviews to assess candidates' genuine thinking abilities, and a research study revealed significant gender disparities in coding agent usage among social science researchers.

Key details:

• Anthropic's hiring process runs candidates through up to five rounds testing skills, values, and ethical thinking, with salaries reaching up to $850,000
• Some applicants pay $4,600 for interview prep coaching run anonymously by current Anthropic employees
• An Anthropic research study found that researchers with typically male names use coding agents more than twice as often as those with typically female names, even within the same discipline and career level
• Usage varies significantly by field: economists lead at 39 percent adoption, while education researchers sit at just 4 percent
• The gender gap for coding agents is far wider than for general AI use

Why it matters: These findings highlight the expanding influence of AI agents in knowledge work while raising questions about equitable adoption. Anthropic's hiring stance signals concern about AI-augmented problem-solving masking genuine individual capability, relevant as AI agents become embedded in professional workflows across industries.

Practical takeaway: If you're interviewing at frontier AI labs, prepare to solve problems without AI assistance; simultaneously, monitor whether your field or team shows disproportionate gendered adoption of coding agents, as this gap may signal training, confidence, or access disparities.

What happened: A large-scale study found that the training process that makes language models into helpful chatbots simultaneously weakens their ability to accurately simulate human behavior.

Key details:

• The study covered 208,000 participants and 26 million responses
• The effect of helpfulness training on behavioral realism worsens with each successive model generation
• Adding demographic profiles (the "persona trick") to improve behavioral accuracy provides practically no benefit for individual predictions
• This represents a fundamental trade-off in how reinforcement learning from human feedback shapes model capabilities

Why it matters: This finding has implications for researchers using AI models to simulate human decision-making, survey responses, or behavioral patterns. It demonstrates that optimization for helpfulness — the dominant training objective across frontier models — comes at the cost of behavioral authenticity, a concern for anyone using models as proxies for human judgment.

Practical takeaway: If you're using AI models to simulate or predict human behavior (for research, testing, or design), be aware that modern RLHF-trained models may not replicate actual human patterns accurately; consider collecting real human data rather than relying on model-generated behavioral simulations.

What happened: SoftBank announced a major European AI infrastructure investment, planning to build AI data centers with up to 5 gigawatts of capacity in France at a cost of up to €75 billion.

Key details:

• This represents SoftBank's largest AI infrastructure investment in Europe
• By 2031, facilities worth €45 billion are planned to go online at three sites in northern France
• The investment reflects the company's strategy of mega-scale AI infrastructure announcements
• SoftBank has a track record of announcing ambitious projects worldwide, though completion timelines vary

Why it matters: This investment underscores the massive capital requirements for frontier AI compute and signals confidence in European AI demand. The 5-gigawatt capacity target represents enormous computational infrastructure, reflecting the infrastructure arms race driven by training and inference demands of large language models.

Practical takeaway: Monitor whether SoftBank's French data center projects come to completion on schedule, as the company's execution history on such megaprojects varies; this investment will shape Europe's AI compute landscape if realized.

What happened: Microsoft and Nvidia are partnering to develop AI PCs powered by Nvidia's own chips as the main processor, moving away from the Copilot+ PC model and toward local agent execution.

Key details:

• The first Windows computers from Dell and Microsoft's Surface line are set to be unveiled at Computex and Build conferences next week
• Nvidia is pushing into the PC market with its own chips as the primary processor, not secondary accelerators
• Microsoft is planning new software likely based on the OpenClaw framework that enables AI agents to handle tasks locally on Windows PCs
• This represents a second attempt at AI PC strategy after the Copilot+ PC concept largely failed

Why it matters: This signals a strategic shift from cloud-dependent Copilot features toward local, agent-based execution on consumer hardware. The partnership between two major computing companies represents a significant bet on locally-run autonomous agents becoming the dominant AI paradigm for PCs, contrasting with browser-based and cloud-first approaches.

Practical takeaway: Expect AI PC announcements at Computex and Build conferences in early June; evaluate whether local agent execution addresses your workflow needs better than cloud-based alternatives.

What happened: Mathematician Terence Tao described how AI could fundamentally reshape mathematical research by enabling true division of labor for the first time in the field's history.

Key details:

• Historically, mathematicians have had to master every step of research from framing problems to verifying results themselves
• Tao envisions "industrial mathematics" emerging: large AI-supported teams rather than individual geniuses
• Humans will remain indispensable in this model, particularly for "inspired guesses" and creative problem framing
• This would represent a structural shift in how mathematical research is organized and conducted

Why it matters: This perspective from one of the world's leading mathematicians suggests AI's potential impact extends beyond tool use into fundamentally reshaping how intellectual work is organized. It echoes historical labor transformations in manufacturing and offers a vision where AI enables specialization and team-based discovery in traditionally individual endeavors.

Practical takeaway: Mathematical researchers and related fields should consider how team-based, AI-augmented workflows might reshape your field; the specialization model Tao describes could create new roles for both human mathematicians and AI systems in discovery.

What happened: Scammers are using AI to generate fake Black personas to deceive social media audiences and sell low-quality dropshipped products, exploiting both platform users and marginalized communities.

Key details:

• The scheme involves creating AI-generated profiles of Black women and other individuals who purport to handcraft or curate products
• These fake personas are used to market products through TikTok Shop and similar platforms, often dropshipped goods from suppliers like Shein
• The deception exploits social and demographic targeting to bypass consumer skepticism
• The practice represents a fusion of AI generation technology with existing e-commerce fraud patterns

Why it matters: This reveals a dangerous convergence of AI-generated content tools and fraud schemes, disproportionately targeting and affecting Black consumers and creators. The use of AI-generated personas with specific racial identities raises serious concerns about AI-enabled discrimination and the erosion of consumer trust in e-commerce platforms.

Practical takeaway: When evaluating social media vendors, especially on TikTok Shop and similar platforms, be skeptical of profiles with limited history or unusual activity patterns; report suspected AI-generated fraud accounts to platforms and exercise extra caution with vendors claiming handmade or curated goods.