AI This Week

DARPA has launched its “Exponentiating Mathematics” (expMath) initiative, a bold move to accelerate mathematical discovery using AI. The goal: build AI “co-authors” — systems capable of breaking down complex proofs into manageable lemmas and composing novel theorems. DARPA is seeking proposals combining large language models, reinforcement learning, and program synthesis to advance symbolic reasoning in mathematics. The roadmap is modular: one track targets AI development itself, while another evaluates AI against professional-level math challenges. If successful, expMath could radically shorten research timelines, impacting cryptography, materials science, fluid dynamics, and more. The program also seeks to democratize access to advanced math tools. DARPA’s ambition is clear: redefine how humans and machines collaborate on deep mathematical inquiry.

A recent NBC News poll finds American views on artificial intelligence echo the early buzz about the internet from the 1990s. AI chat tools such as ChatGPT, Google’s Gemini and Microsoft’s Copilot become widely adopted. Nearly three-fourths of adults report having used these tools while 44% say they use them regularly. The poll draws parallels with past attitudes toward the internet, noting that many concerns remain as new challenges and benefits arise. Opinions sharply divide on AI’s promise; 44% believe AI will improve life, while 42% worry it will harm society. Many even support classroom bans as a way to better prepare students for the future.

Reddit has unveiled two AI-powered ad tools — Reddit Insights and Conversation Summary Add‑ons — built on its “Community Intelligence” engine. These tools analyze billions of comments and posts to extract real‑time trends and highlight positive user feedback directly beneath sponsored content. They aim to help advertisers craft smarter campaigns and deepen engagement by weaving authentic community voices into ads. Launched at Cannes Lions Festival 2025, they reflect a broader industry shift toward AI-driven marketing amid sluggish ad markets. The move also raises privacy questions, as users’ public input fuels ad targeting without explicit notice. These features mark a major push by Reddit to monetize its vast user-generated content while stepping into next‑gen advertising tactics.

TikTok is introducing new artificial intelligence-powered advertising features. The platform now allows marketers to generate video ads using text prompts or still images. Advertisers can upload product images or provide text descriptions, and TikTok's AI tools will create various five-second video clips suitable for ads. This move expands TikTok's AI-generated content capabilities, offering advertisers more options to create engaging video ads quickly and efficiently. The new tools aim to streamline the ad creation process and enhance marketers' ability to produce content tailored for the platform.

Scale AI confirms that Meta has made a landmark minority investment — $14.3 billion for a 49% stake — valuing the AI‑data firm at over $29 billion. As part of the deal, Scale founder and CEO Alexandr Wang is stepping down to join Meta’s “superintelligence” lab, while remaining on Scale’s board. Jason Droege, formerly the company’s chief strategy officer, assumes the role of interim CEO. Scale will use the fresh capital to pump up R&D, deepen partnerships, and pay out shareholders and equity‑holders. Meta gains privileged access to Scale’s AI‑training data network and executive talent amid its push to revive its AI efforts following lukewarm receptions to LLaMA 4.

Online publishers are seeing search referrals implode as Google swaps blue links for AI-generated summaries. The WSJ reports that features like AI Overviews and the new conversational AI Mode are siphoning organic traffic — Business Insider slumped 55% since 2022, HuffPost and The Washington Post face half their search traffic lost — forcing layoffs and strategy pivots. Google claims AI tools boost engagement and still promote links, but publishers say the shift turns Google into an “answer engine,” threatening journalism and prompting deals or lawsuits over content rights

Mary Meeker, known as the "Queen of the Internet," released a report highlighting the unprecedented pace of artificial intelligence adoption. The study shows AI is being invested in, developed, and used faster than any previous technological change. ChatGPT reached 100 million users in under two months, far quicker than Facebook's 4.5 years. Meeker attributes this rapid growth to aggressive development by tech companies, decreased costs, and widespread buy-in from businesses. The report also notes increased mentions of AI in corporate earnings reports and a 167% year-over-year increase in new AI models since 2020.

Vanderbilt University Medical Center researchers have developed two innovative AI technologies to improve cancer care. The first, MSI-SEER, predicts microsatellite instability-high status from pathology slides, potentially identifying more patients who could benefit from immunotherapy. The second is a 3D imaging tool that provides detailed tissue analysis at the subcellular level. These advancements demonstrate how VUMC is using AI to address various medical needs. The 3D imaging technology has applications beyond cancer diagnostics, including drug development and organ transplant assessment.

ChatGPT, the wildly popular AI chatbot, faced a significant outage on Tuesday, leaving users frustrated and unable to access its conversational capabilities. Over 1,500 reports flooded Downdetector, as the platform displayed error messages and failed to respond to queries. The disruption comes on the heels of OpenAI's announcement of reaching $10 billion in annual recurring revenue, a staggering leap from last year's $5.5 billion. Despite the impressive growth, the company reported a substantial loss of approximately $5 billion in the previous year. The outage coincides with the recent release of Advanced Voice, an update for paid users promising enhanced intonation and more natural interactions.

Scientists at Berlin's Charité University Hospital have developed an AI model capable of identifying over 170 tumor types from all human organs. The model, called crossNN, focuses particularly on difficult-to-diagnose brain tumors. It analyzes epigenetic characteristics of tumors, providing highly accurate diagnoses without the need for invasive surgery in some cases. The AI achieves 99.1% accuracy for brain tumors and 97.8% accuracy across all tumor types. Researchers plan to conduct clinical trials across Germany to further test crossNN's accuracy before potentially introducing it into routine cancer care.