GoKawiil
Tech News
A new web tool tested 12 AI models by asking them to rank top three choices on various topics. The results revealed that four Chinese models—DeepSeek, Kimi K2, Tencent’s Hy3, and GLM—tended to produce answers more consistent with the overall group consensus. The scoring system measured each model's agreement with others, highlighting differences in conformity among the models.
The Chan Zuckerberg Biohub announced a partnership with the Department of Energy and the National Institutes of Health to invest $1.8 billion in generating biological datasets for AI research. Biohub stated the funding is meant to give researchers data that AI models can use to identify new approaches to preventing and treating diseases.
French AI company Mistral has launched a preview of Mistral Large 4, a 1 trillion-parameter open-weight model nicknamed Le Chonk, with a final version due by month's end. The model is freely usable and customizable, and Mistral says it was trained from scratch with particular optimization for coding, cyberdefense, manufacturing, finance and electrical engineering tasks.
Researchers released OpenWAM, an open framework that lets developers compare and combine different 'world-action' approaches for robots—systems that predict future video, generate actions, or do both together. The team pretrained a shared video-action model, adapted from Wan2.2-5B, on roughly 3.34 million robot and human interaction trajectories totaling 14.64k hours of video, training it for 14 days on 32 NVIDIA B200 GPUs using video data alone before adding action learning.
Strands Labs has launched Strands Decider 2B, a small open-source 'decision model' built for fast local experimentation within its strands-labs initiative. Unlike general-purpose LLMs, it is designed to choose among predefined options or assign numerical scores rather than generate free-form text, and it also outputs a reliability score for each decision. The model is intended to integrate with the Strands Harness SDK and the recently launched Strands harness.
A study published in Nature by Minixhofer and colleagues introduces 'byteification,' a method for converting existing token-based large language models so they can process text as individual bytes rather than whole word fragments. The authors report that byteified models retain the ability to identify individual characters within words while still matching the performance of standard token-based models.
Musubi unveiled PolicyLM-1.7B, a lightweight decision model built to moderate content in under 50 milliseconds using plain-English policies, and released it with open weights. The company says it matches the cost and speed of existing moderation classifiers but can adapt to new rules without retraining, unlike conventional systems.
Linum's research team has published Pyramid-JiT (P-JiT), a new decoder-only pixel-space architecture that predicts images at multiple resolutions through the diffusion transformer trunk, building on their earlier VAE-free JiT-DDT model. The company says P-JiT matches the final FD-DINOv2 score of its Linum v2 model using 11.3 times fewer training samples and 4.3 times fewer GPU-hours, while operating at four times the pixel resolution. Code and weights have been released under an Apache 2.0 license as a research artifact rather than a finished product.
Mistral AI has released a public preview of Mistral Large 4, its new flagship open-weight model nicknamed 'le Chonk.' The company says it processes both text and images, improves on cybersecurity and agentic task performance, and reaches state-of-the-art results on benchmarks like SWE-bench and Terminal Bench.
OpenAI reportedly plans to release on GitHub hundreds of mathematical solutions generated by an internal model it began training in August, including a claimed resolution of the Navier-Stokes Millennium Prize problem. Mathematicians who attended an August meeting with OpenAI say they asked the company to publish such findings as peer-reviewable papers rather than a bulk dump, a request they say was ignored. OpenAI spokesperson Lindsay McCallum says the company is consulting an advisory group at the Institute for Advanced Study and has not set a release date.
French AI startup Mistral released a new open-weight model, which it says performs competitively against leading open-source models developed in China. The company positioned the release as part of its effort to offer a European alternative in the global open-source AI race.
Mirror Particle, a two-year-old San Francisco startup, is developing a custom foundation 'world model' that aims to predict consumer behavior and its underlying causes, rather than relying on large language models fine-tuned to role-play demographics. CEO Abhivyakti Ahuja says the model focuses on tracking how people change over time, including the triggers behind those shifts. The company has raised an angel round and says it is close to closing its first venture round.