Import AI 457 focuses on three studies. SentinelOne analyzed fast16.sys, a virus from more than 20 years ago that corrupted results from high-precision software such as LS-DYNA 970, PKPM, and MOHID by tampering with in-memory calculations, sabotaging scientific and engineering work.
University of Cambridge professor Clare Bryant used Google DeepMind's Co-Scientist to study how pathogens such as influenza trigger sepsis when transmitted across species. Among the hypotheses the tool generated and ranked was a protein she had not previously considered, leading the research to a specific amino acid. Bryant says experiments that would normally take two to three years to identify the precise amino acid could now be completed in six months. Her team is building cell lines with amino-acid mutations to test the hypothesis.
Matt Onsum and Katherine Labbé of Calico Life Sciences are using Google DeepMind's Co-Scientist to integrate scattered findings in aging biology and generate testable hypotheses.
University of Edinburgh bioengineer Filippo Menolascina used Google Co-Scientist to study MASH liver disease. The system integrated evidence from liver biology and pharmacology, identified mechanisms worth investigating, and flagged candidate combination therapies for testing.
Google's Co-Scientist helped MIT mechanical engineer Ritu Raman rapidly review ALS literature, turn ideas into testable hypotheses, and rank them by feasibility. Its strongest lead pointed to interactions between molecules on cell surfaces, outside Raman's field, prompting her to work with Boston Children's Hospital chemical biologist Ryan Flynn. Together, using their respective toolkits and iterating with Co-Scientist, they are searching for new RNA mechanisms—and potentially RNA drugs—that could target ALS.
Stanford School of Medicine geneticist Gary Peltz used Google DeepMind's Co-Scientist to screen for anti-fibrotic drugs. Of the three candidates it proposed, two successfully blocked fibrosis and promoted liver-cell regeneration in experiments with live human liver cells; neither of the two candidates Peltz selected himself was effective.
In Import AI 455, Jack Clark estimates a greater than 60% chance that by the end of 2028 AI R&D will run without human involvement, with frontier models autonomously training their own successors. He puts the chance of this happening by 2027 at about 30%.
Google Research says its open-source tools and datasets have enabled more than 250,000 researchers and developers worldwide across fields including genomics, neuroscience, and Earth modeling. Tools such as DeepVariant have supported analysis of whole-genome and exome data from 2.5 million people, while the HAI-DEF model suite, including MedGemma, has been downloaded more than 4.8 million times.
Google Research summarizes four areas where scientists use Empirical Research Assistance (ERA): in public health, the team submits weekly influenza forecasts for U.S. states.
Why it matters: Google uses four real research projects to show ERA deployments in epidemic forecasting, cosmology, carbon monitoring, and neuroscience, illustrating how AI-assisted research is being used.
Google DeepMind and South Korea's Ministry of Science and ICT (MSIT) have announced a partnership that will establish an AI Campus at Google DeepMind's Seoul office and give South Korean academia access to frontier models including AlphaEvolve, AlphaGenome, AlphaFold, AI co-scientist, and WeatherNext.
Google DeepMind has released Decoupled DiLoCo, a distributed-training architecture that divides large-scale training into compute islands communicating asynchronously, reducing bandwidth requirements and isolating hardware failures. In tests, training continued and reconnected after an entire learner unit went offline, achieving the same ML performance as traditional methods with Gemma 4 models.
Why it matters: The source reports distributed pretraining results across bandwidth, fault tolerance, and mixed-generation hardware, helping readers assess engineering tradeoffs.
A Berkeley AI Research team has proposed GRASP, a gradient-based planner that makes long-horizon planning with learned dynamics (world models) feasible. It lifts trajectories into virtual states for parallel optimization across time, injects stochasticity directly during iterative state refinement to explore, and reshapes gradients so actions receive clearer signals. The method addresses ill-conditioned computation graphs, local minima caused by non-greedy structure, and fragile gradients in high-dimensional visual models.
Google Research has published “Reasoning-Driven Synthetic Data Generation and Evaluation,” proposing Simula, a framework that treats synthetic-data generation as a mechanism-design problem. A reasoning model builds an entire dataset from scratch without seed data.
Google Research has proposed MoGen (Neuronal Morphology Generation), a model that generates synthetic neuron shapes with point-cloud flow matching. The data supplements training for the PATHFINDER reconstruction model, reducing mouse-axon reconstruction error by 4.4%, mainly by decreasing merge errors. At whole-mouse-brain scale, the work is equivalent to saving 157 person-years of manual proofreading. MoGen is available as an open-source model.
Google Research has introduced ConvApparel, a fashion-shopping dataset containing more than 4,000 multi-turn human–AI conversations (nearly 15,000 turns) for measuring the realism gap between LLM user simulators and real human behavior.
In the paper “Forest vs Tree: The (N, K) Trade-off in Reproducible ML Evaluation,” Google Research studies the tradeoff between the number of evaluation items and annotators per item. It finds that the commonly used one to five annotators per item is often insufficient; more than ten may be needed to capture variation in human opinion.
Google Research has released S2Vec, a self-supervised framework that converts the built environment into layered images through S2 Geometry tiling and feature rasterization, then learns general-purpose embeddings with a masked autoencoder (MAE). On zero-shot geographic extrapolation for U.S.-wide socioeconomic predictions such as population density and median income, S2Vec is usually the best single model. Multimodal fusion with image embeddings generally performs better than either modality alone, though tasks such as tree canopy and elevation remain challenging.
Two companion studies in Nature Cancer evaluate an AI breast-screening system in the AIMS research collaboration between Google Research and several NHS organizations.
Why it matters: The two Nature Cancer studies report detection-rate and workload results for AI in the NHS double-reading workflow, showing the practical limits of human–AI collaboration.
Mistral AI has released Forge, a system that lets enterprises build frontier-level AI models from their own proprietary knowledge. It spans pretraining, post-training, and reinforcement learning; supports dense and MoE architectures; and accepts multimodal inputs.
Why it matters: The source details the stages of training frontier models on enterprise data and names organizations already involved, helping readers assess the path and barriers to building models in-house.
Berkeley AI Research has proposed SPEX and ProxySPEX, algorithms that identify important interactions driving LLM outputs across feature, data, and model-component attribution. SPEX turns interaction search into a sparse-recovery problem by exploiting sparsity and low-order structure; ProxySPEX uses additional hierarchy to match SPEX performance with about ten times fewer ablation experiments.
Google has announced that Flood Hub now includes urban flash-flood forecasts, using a new AI approach to predict urban flood risk up to 24 hours in advance.
Why it matters: The source describes the modeling approach and evaluation results for urban flash-flood forecasting, helping readers assess the reach of AI flood warnings in the Global South.
Google has released Groundsource, a method for using Gemini to extract structured disaster data from global news. It has also opened the first urban flash-flood dataset, with 2.6 million records across more than 150 countries from 2000 to the present.
Why it matters: Google turns global news into structured disaster data with Gemini and reports human-verified accuracy and coverage comparisons, helping readers assess whether the method can extend to other disasters.
A Berkeley AI Research team proposes a mutual-information-based framework for evaluating and optimizing imaging systems. It estimates information directly from noisy measurements to predict system performance. Validated in color photography, radio astronomy, lensless imaging, and microscopy, optimized designs reach end-to-end-method performance with lower memory and compute needs and no task-specific decoder. The paper appeared at NeurIPS 2025.
Berkeley AI Research has proposed a divide-and-conquer off-policy reinforcement-learning algorithm that does not rely on temporal-difference (TD) learning. It reduces the number of Bellman-recursion steps from linear to logarithmic, enabling scaling to long-horizon tasks. The method achieves scaling for the first time in goal-conditioned RL, using the triangle inequality for shortest-path distances between states to construct transitive Bellman updates. Selecting the optimal subgoal w remains an open problem.
Berkeley AI Research proves that under a fourth-order approximate objective, specific hyperparameter constraints, tiny initial embedding weights, and tiny gradient-descent steps, word2vec's learning problem reduces to an unweighted least-squares matrix factorization. Its gradient-flow dynamics have a closed-form solution, and the resulting representations are equivalent to applying PCA to matrix M*.
Mistral fine-tuned Pixtral-12B with LoRA for satellite-image classification on the Aerial Image Dataset (AID), substantially outperforming the untuned baseline and improving discrimination between easily confused categories such as dense versus medium residential areas and playgrounds versus stadiums.
Mistral AI, Carbone 4, and the French Agency for Ecological Transition (ADEME) completed the first full lifecycle analysis of an AI model and disclosed the environmental footprint of Mistral Large 2: by January 2025, after 18 months of use, it had produced 20.4 ktCO₂e, consumed 281,000 m³ of water, and used 660 kg Sb eq.
