323,492 AI headlines dropped this week. here were the power moves that everyone needs to know about: Meta dropped $300M to build its own AI Avengers, ransacking OpenAI’s payroll in the process. Sam Altman clapped back—purpose over profits—and launched a $10M+ enterprise consulting arm to prove it. Meanwhile, medical superintelligence isn’t science fiction anymore. Microsoft’s AI doctor outperformed real ones. Chai-2 cracked biotech timelines. Sakana open-sourced the era of model teamwork. The shift is clear: it's not about who has the biggest model. It's about who can build the smartest system—and the sharpest mission. Let’s dive in... Subscribe for more weekly reviews like these! Altman claps back at Meta’s recruiting surge The News: Sam Altman addressed OpenAI researchers Monday night in a fiery internal message, calling Meta’s recent recruitment tactics "distasteful" and warning of potential "very deep cultural problems" at Meta. The details: • Meta reportedly offered up to $300M over four years to top talent but failed to secure OpenAI’s "top people," instead hiring several notable researchers including Trapit Bansal, Shuchao Bi, Hongyu Ren, Jiahui Yu, and Shengjia Zhao. • Altman reassured staff that OpenAI’s compensation structures are being reviewed, arguing that its stock offers "much, much more upside" than Meta’s. • He contrasted OpenAI’s mission-driven culture with what he called Meta’s "flavor of the week" mentality. • Meta's new Superintelligence Labs have already onboarded 11 researchers from OpenAI, Google, and Anthropic. Why it matters: OpenAI CRO Mark Chen compared Meta’s poaching to "someone breaking into our home," capturing the emotional toll. Altman remains confident OpenAI’s values will retain top talent and win the long game, even amid an escalating war for AI minds. Sakana AI enables model teamwork The News: Sakana AI, a Tokyo-based lab, has introduced AB-MCTS (Adaptive Branching Monte Carlo Tree Search), an algorithm that enables advanced AI models like ChatGPT, Gemini, and DeepSeek to collaborate on complex tasks — shifting away from the "bigger is better" approach in AI. The Details: • AB-MCTS extends the Monte Carlo Tree Search method from AlphaGo to inference time, letting the system choose whether to refine a solution or generate new ones at each step. • The algorithm selects the most appropriate model for each sub-task: one might handle logical reasoning, another excels at code generation. See implementation details. • If one model errs, another can identify and correct the output — leading to better performance overall. • On the ARC-AGI-2 benchmark, the team’s multi-agent setup solved 30% of problems, outperforming top solo models (23%) by 7%. • The underlying framework, TreeQuest, is now open source under Apache 2.0. Why It Matters: AB-MCTS is ushering in an era of inference-time scaling and model cooperation. Just as human brains rely on specialized regions working together, this architecture points to a future where AI agents combine strengths to surpass monolithic models — accelerating our path to AGI. Chai-2: AI generates breakthrough antibodies The News: OpenAI-backed Chai Discovery released Chai-2, an AI system that designs functional antibodies with nearly a 20% success rate — a 100x leap over conventional methods. The details: • Chai-2 generated antibody candidates for 52 disease targets, achieving validated outcomes in almost half after testing just 20 per target — a process that typically requires screening millions. • Traditional antibody discovery takes months or even years, but Chai-2 delivers results in just two weeks. • The system designs antibodies from scratch using only the target’s molecular structure, without needing prior antibody examples. • Often described as "Photoshop for proteins", it gives researchers granular control over where antibodies bind to disease targets. Why it matters: The bottleneck in modern medicine isn’t scientific capability — it’s time and cost. AI tools like Chai-2 have the potential to democratize precision medicine, especially for rare diseases that often go untreated due to high R&D barriers. Microsoft’s move toward medical super intelligence: The News: Microsoft just rolled out the MAI Diagnostic Orchestrator (MAI-DxO), an AI tool outperforming seasoned physicians in diagnosing complex medical cases, marking what its team calls a "step toward medical super intelligence." The details: • MAI-DxO simulates a virtual panel of clinicians by orchestrating five specialized AI agents, each emulating different medical specialties such as hypothesis generation, test prioritization, and cost oversight. • It was tested against SDBench, a benchmark based on 304 complex cases from the New England Journal of Medicine, achieving a diagnostic accuracy of 85.5% using a blend of LLMs, including OpenAI's o3. • In contrast, 21 experienced U.S. and U.K. physicians, operating without peer consultation or references, averaged just 20% accuracy. • MAI-DxO also optimized healthcare spending, reducing diagnostic costs to $2,397 per case compared to $2,963 for human-led cases. Why it matters: By integrating top-tier language models like o3, Gemini, LLaMA, and Grok in a collaborative framework, MAI-DxO offers a new model for scalable medical expertise. Its ability to cut unnecessary testing while solving high-complexity diagnoses reframes how AI can support physicians — not replace them — in delivering better, faster, and more cost-effective care.’ What an 'AI Manhattan Project' could really look like: The News: Epoch AI released a detailed analysis of what a modern, U.S.-led "AI Manhattan Project" might involve. Echoing calls from the U.S.-China Economic & Security Review Commission, this blueprint models the initiative after historic efforts like Apollo and the original Manhattan Project. Details: • The proposal suggests coordinating all national compute under one agency — enabling the U.S. to leap ahead in AI scaling with over 27 million GPUs. • Total investment could range between $100–$244B per year, or ~0.4–0.8% of U.S. GDP. • Infrastructure build out would require invoking the Defense Production Act to construct power plants, data centers, and critical supply chain assets. • Analysts suggest pairing short-term solutions (natural gas turbines) with longer-term plays like small modular reactors. • Security concerns loom large — from espionage to AI misuse — prompting debate over public vs. classified approaches. Why it matters: As global AI competition heats up, building national AI infrastructure is no longer optional. A unified national effort could accelerate AGI timelines by years, shift economic and defense power balances, and reshape the very architecture of innovation — but only if the U.S. can mobilize fast enough to outpace rivals like China. Thanks for reading this far! Stay ahead of the curve with my daily AI newsletter—bringing you the latest in AI news, innovation, and leadership every single day, 365 days a year. See you tomorrow for more!