Interconnects
·
2 months ago
Multiple open-source AI labs released new frontier models this month, including DeepSeek V4, Gemma 4, Kimi K2.6, and others, prompting a fresh evaluation of how open models compare to closed alternatives. The Center for AI Standards and Innovation found open models lag by 3-7 months behind American frontier models, with the gap widening when measured against proprietary benchmarks like CTF-Archive-Diamond and PortBench. The benchmarking methodology itself remains contested, as standardized evaluation setups do not match how models are actually deployed—for example, using basic coding harnesses rather than the specialized environments models are trained for.
Ahead of AI
·
2 months ago
● 2 sources
Recent open-weight LLM releases including Gemma 4, DeepSeek V4, and others have adopted architectural techniques like KV sharing across layers, per-layer embeddings, and compressed attention to reduce memory and compute costs for long-context processing. Gemma 4 E2B achieves approximately 2.7 GB of KV cache savings at 128K context length through cross-layer KV sharing that allows later transformer layers to reuse key-value tensors from earlier layers. These efficiency-focused design changes enable smaller models to handle longer contexts and reduce memory requirements, which becomes critical as reasoning models and agent workflows maintain more tokens during inference.
Google DeepMind
·
2 months ago
Google DeepMind is launching new programmes in Singapore as part of a national partnership with the Singapore Government to apply frontier AI to healthcare, education, and scientific research. The partnership aims to create an additional S$3.3 billion in economic value through faster R&D by 2040, with initial applications including AI co-clinician research, pandemic preparedness initiatives, and Gemini access for all educators from primary schools to junior colleges. The expanded presence will support public sector transformation, workforce development, and responsible AI deployment across the Asia-Pacific region.
Google DeepMind
·
2 months ago
● 3 sources
Researchers at Cambridge are using an AI tool called Co-Scientist to identify molecular mechanisms that cause severe disease when pathogens jump from animals to humans. Professor Clare Bryant's lab tested the tool on flu research and found it could narrow experimental focus from candidate proteins down to specific amino acids in six months instead of the typical two to three years. If the AI-guided approach identifies the correct targets, this could accelerate the discovery of intervention points to prevent zoonotic disease severity in humans.
Google DeepMind
·
2 months ago
● 2 sources
Calico Life Sciences used an AI tool called Co-Scientist to analyze aging research literature and generate testable hypotheses about how cells protect themselves against stress. The team applied this approach to study the integrated stress response, a cellular mechanism that changes with age, and designed experiments based on AI-generated hypotheses that produced novel findings. The results will be published and may inform understanding of how cellular stress response contributes to aging-related diseases.
Google DeepMind
·
2 months ago
● 3 sources
A biomedical team at the University of Edinburgh used an AI system called Co-Scientist to identify overlooked biological connections in metabolic dysfunction-associated steatohepatitis (MASH), a liver disease involving multiple intertwined processes. The AI synthesised evidence across liver biology and pharmacology to propose that the NLRP3 inflammasome links inflammation and metabolism in MASH, a connection that had not previously been integrated into a single explanation. This hypothesis, later experimentally verified, could enable the development of targeted combination therapies for patients who do not respond to existing single-target drugs like resmetirom.
Google DeepMind
·
2 months ago
Ritu Raman and Ryan Flynn are collaborating on ALS research by combining their separate expertise in tissue engineering and RNA biology, facilitated by an AI tool called Co-Scientist. Co-Scientist helped Raman rapidly synthesize months of fragmented ALS literature and generate testable hypotheses for her tissue models within weeks. Their combined approach now targets RNA-based mechanisms at cell surfaces as potential therapeutic avenues for ALS.
Google DeepMind
·
2 months ago
● 3 sources
Researchers at Stanford University used an AI tool called Co-Scientist to identify existing medicines that could treat liver fibrosis, a scarring condition that kills over 1.4 million people annually. When tested against human liver cells, two of three AI-selected drug candidates blocked fibrosis progression, while the cancer drug vorinostat blocked 91% of a damage response driving liver scarring. The findings suggest that drugs altering gene activity patterns warrant clinical investigation as potential treatments for the condition.
Google DeepMind
·
2 months ago
The National Hurricane Center used Google DeepMind's WeatherNext AI model to predict Hurricane Melissa would reach Category 5 strength in Jamaica five days in advance with 80% confidence, marking the first time a storm was successfully predicted to reach that intensity from such weak initial conditions. The five-day lead time allowed Jamaica's Meteorological Service to coordinate evacuations and resource mobilization before the October 2025 landfall. Early warnings enabled officials to reduce harm to communities and protect livelihoods that would otherwise have been lost.
OpenAI Blog
·
2 months ago
OpenAI and Malta have partnered to provide ChatGPT Plus access to all citizens along with training programs on responsible AI use. The initiative includes educational resources designed to help Maltese residents develop practical skills for working with AI tools. The partnership aims to increase digital literacy and ensure broader access to advanced AI capabilities across the population.
