GoKawiil
Tech News
Danijar Hafner, a 31-year-old AI researcher named to MIT Technology Review's 35 Innovators Under 35 list, runs a quiet San Francisco startup filled with humanoid robots suspended from racks. He has spent years refining world-model-based agents by testing them against video games, and is now moving that work from simulations into physical robots operating in real environments.
U.S. security agencies have publicly accused six Chinese artificial intelligence companies of systematically using American AI models to train their own systems through a technique known as distillation. The practice allows a smaller model to learn from the outputs of a larger, more advanced one, effectively transferring capabilities without the original research investment.
An Oxford University researcher who studies AI's effect on decision-making argues that generative AI tools should be treated as intellectual sparring partners rather than sources of ready-made answers. The piece cites data showing 94% of UK undergraduates use AI for assessed coursework and that over half of 2025 scientific papers show signs of LLM-generated text, up from roughly one in ten in 2023. The author contends fears about AI eroding critical thinking are misplaced, and that the real risk lies in passive use rather than active engagement with the technology.
Researchers introduced GPN-Star, a phylogeny-informed genomic language model that outperforms established tools like PhastCons and PhyloP in identifying functional constraint and deleterious variants across the human genome. The model showed especially strong results for distal enhancers, a class of regulatory DNA that has been notoriously hard to analyze, and it learned meaningful patterns linking enhancers, transcription factor binding sites, and their co-evolution without needing labeled training data.
The provided text is a bibliography of academic papers and preprints on computational and AI-based approaches to modeling cells, including foundation models for single-cell biology, perturbation prediction, and optimal transport methods for tracking cellular dynamics. It compiles work from research groups building toward what scientists call a 'virtual cell'—computational systems that simulate how cells respond to genetic or chemical changes. No new announcement, product, or specific finding is presented; it is purely a citation list.
Researchers identified a class of long-range GABAergic (Sst+/Nos1+) inhibitory neurons in the mouse neocortex that project across distant cortical regions rather than staying local. Using genetic mouse models, in vivo imaging and chemogenetic manipulation, they showed these neurons drive widespread cortical synchrony and are linked to sleep regulation.
Google DeepMind's Demis Hassabis proposed training a large language model only on scientific knowledge available before 1911 to see if it could independently derive general relativity, calling it a meaningful test for artificial general intelligence. Several research teams, including one led by Ido Kaminer at Technion, have since built such 'vintage' AI models restricted to historical data, but early results have exposed shortcomings in current systems rather than demonstrating breakthrough reasoning.
An Anthropic employee has left the company, warning that competitive pressure among AI labs is pushing them to build self-improving models that could escape human oversight. The departure reflects growing internal unease about the pace of AI capability development outstripping safety safeguards.
Researchers report that large language models can develop novel social biases not present in their original training data when placed in adaptive, exploration-based learning settings. These biases emerged as the models optimized for rewards through repeated interactions, effectively creating stereotype-like associations on their own.
Google has released version 3 of its WeatherNext AI weather forecasting model, which now incorporates raw satellite data instead of relying solely on six-hourly reanalysis snapshots used by most AI weather systems. The update allows the model to generate hourly forecasts instead of every six hours, while also increasing spatial resolution through a larger machine-learning model, according to a newly published white paper.
An international team of scientists published a not-yet-peer-reviewed paper arguing that large language models like ChatGPT and Claude spread through populations similarly to how viruses transmit, driven by social learning and usefulness rather than biology. They contend this spread encourages 'cognitive offloading,' where people increasingly hand over mental tasks like reasoning and writing to AI tools, deepening reliance over time.
Motif Design has launched a browser-based platform that lets architecture teams work with highly complex building models directly online, without specialized desktop software. The tool is positioned as a collaboration-first alternative to traditional architectural modeling workflows, echoing the browser-native approach that made Figma popular in design.