Highlights
Top Insights
1. Leading biopharma companies have cut early drug discovery timelines from 4–5 years to ~8 months, a scale of acceleration that far exceeds typical productivity gains.
2. While most companies run AI pilots, very few treat AI as the “central nervous system” of the organization. That mindset shift, not technology, is the real differentiator.
3. AI is reshaping core scientific work, not just admin tasks. AI agents can now orchestrate complex R&D workflows (like trial documentation) 4× faster and at a fraction of the cost, fundamentally changing how science is executed.
Source: The AI-First Biopharma Company (BCG)
Top News
1. Alibaba’s Qwen-Image-2512 provides a powerful, open-source alternative to Gemini 3 Pro Image.
2. DeepSeek, in a paper, introduces a new training-cost-efficient architecture called Manifold-Constrained Hyper-Connections (mHC).
3. Alibaba’s Tongyi Lab has open-sourced MAI-UI, a high-performing GUI agent framework.
4. Meta has agreed to acquire Manus to strengthen its AI capabilities with agent technology.
Additional Insights
1. Science in 2050: the future breakthroughs that will shape our world — and beyond (Nature)
The article outlines a future defined by a shift toward artificial general intelligence (AGI) potentially conducting most scientific research, alongside critical advancements in nuclear fusion and quantum sensors that may finally resolve mysteries regarding dark matter. While climate change is expected to surpass the $2°C$ warming threshold—sparking high-stakes debates over geoengineering and the industrial profitability of carbon capture—medical science is poised to transition into a quantitative era of biomarker-based diagnostics for neurological disorders. However, these technological leaps, including the colonization of Mars and the development of claytronics (programmable matter), face significant hurdles from biological limitations, geopolitical instability, and the potential for populism to erode the public funding essential for long-term discovery.
2. The greatest show on Earth (The Economist)
Sam Altman is leading OpenAI through a high-stakes expansion, aggressively diversifying into custom chips, consumer hardware, and business consulting while facing a staggering financial burn rate expected to hit $17 billion in 2026. Despite achieving unprecedented revenue growth—reaching an annualized rate of $20 billion by the end of 2025—the company remains unprofitable because its massive computing costs remain tethered to its revenue. OpenAI’s dominance is currently under siege by Google’s Gemini 3, which has outperformed GPT-5.1, and by the rising popularity of Anthropic among enterprise users. To sustain its $1.4 trillion long-term infrastructure goals, Altman is seeking a massive valuation of up to $830 billion and new revenue streams like advertising and e-commerce; however, skeptics warn that the company’s “WeWork-like” deficit and reliance on constant fundraising make it a precarious high-wire act that must soon prove it can generate a profit.
3. Tech To Track in 2026 (IEEE Spectrum 1, 2)
A wave of under-the-radar innovations is poised to reshape energy, computing, medicine, transportation, and space: grid-scale energy storage is advancing through novel approaches like CO₂ “bubble batteries” and gravity systems to support AI-hungry data centers; data centers themselves may be transformed by low-power radio-frequency interconnects integrated directly with GPUs; noninvasive ultrasound treatments that use cavitation bubbles show promise against deadly cancers; and drones are expanding from logistics into wildfire detection and suppression. Alongside these are landmark developments such as electric air taxis, autonomous consumer vehicles, foldable iPhones, AI-generated advertising, robo-umpires in professional sports, and massive AI supercomputers consuming gigawatts of power. Space and deep-tech milestones—including a return to crewed lunar missions, asteroid sample returns, in-situ resource extraction for Moon and Mars exploration, advances toward commercial fusion, and pivotal decisions in next-generation chipmaking—underscore 2026 as a year when experimental technologies begin crossing into real-world scale and societal impact.
4. Steam, Steel, and Infinite Minds (Ivan Zhao)
The essay argues that AI is the next “miracle material,” akin to steel in the Gilded Age or steam in the Industrial Revolution, with the power to fundamentally reshape knowledge work, organizations, and economies. Today, we are still using AI through the lens of the past—bolting chatbots onto human-centric workflows—much like early factories merely swapped waterwheels for steam engines. True transformation will come when AI agents, or “infinite minds,” consolidate fragmented context, enable verifiable outcomes, and shift humans from doing work to supervising it at leverage. At the individual level, this moves knowledge workers from “bicycles” to “cars”; at the organizational level, AI becomes structural steel that allows companies to scale without collapsing under communication overhead; and at the economic level, it enables a leap from human-scaled “Florences” to megacity-like knowledge economies that run continuously. The challenge and opportunity lie in abandoning rearview-mirror thinking and redesigning work, companies, and rhythms of decision-making around AI’s fundamentally new capabilities.
Innovation Radar
1. AI Model Releases and Advancements
Liquid AI has released LFM2-2.6B-Exp, an experimental 3B-class model that uses pure reinforcement learning on a hybrid convolution-attention architecture to achieve industry-leading instruction following and math performance for on-device deployment (MarkTechPost).
Tencent has released HY-Motion 1.0, an open-weight, billion-parameter text-to-3D human motion model that uses a Diffusion Transformer (DiT) and Flow Matching to generate high-quality animations from natural language prompts (MarkTechPost).
Naver’s new HyperCLOVA X SEED Think model has debuted on the Artificial Analysis Intelligence Index, ranking as the second-highest Korean model with strong performance in reasoning and agent tool usage (AJU Press).
Alibaba’s Qwen-Image-2512 provides a powerful, open-source alternative to proprietary models like Google’s Gemini 3 Pro Image, offering enterprises high-fidelity image generation and layout control under a flexible Apache 2.0 license (VentureBeat).
IQuest Research has released a series of open-source coding models called IQuest-Coder-V1, featuring a 40B model that reportedly outperforms GPT-5.1 on the SWE-bench benchmark (Sohu).
DeepSeek, in a paper, introduces a new architecture called Manifold-Constrained Hyper-Connections (mHC), designed to enable the training of larger, more scalable AI models with negligible additional computational costs (SCMP).
2. AI Tools and Features
Alibaba’s Tongyi Lab has open-sourced MAI-UI, a high-performing GUI agent framework that uses a hybrid end-cloud system to enhance privacy, streamline cross-app tasks (Pandaily).
Meta has agreed to acquire Chinese-founded AI start-up Manus—reportedly for over $2bn—to strengthen its AI capabilities with autonomous “agent” technology that can independently plan and complete complex tasks (BBC).
3. Others
Researchers have developed a biology-inspired “neuromorphic” artificial skin for robots that uses energy-efficient electrical spikes to communicate pressure, detect pain, and trigger reflexive responses (Ars Technica).
A Washington state man successfully completed the first-ever fully autonomous, coast-to-coast drive across the U.S. using Tesla’s FSD v14.2 software with zero human disengagements (NYPost).
Elon Musk announced that Neuralink will begin high-volume production and automated surgical implementation of its brain-computer interface devices in 2026 (FierceBiotech).
