Alibaba-backed Chinese AI startup Moonshot has unveiled its latest model, Kimi K3, which boasts 2.8 trillion parameters, making it potentially the largest open-weight model released to date. This release is causing significant disruption in the industry, with some benchmarks showing Kimi K3 outperforming leading models from Anthropic and OpenAI.

While Moonshot acknowledges that K3's overall performance still trails Claude Fable 5 and GPT-5.6 Sol, independent testing by Artificial Analysis places it closely behind these proprietary systems. Notably, on Arena.ai's front-end development leaderboard, Kimi K3 ranks above the two most powerful models, a 17-place jump from its predecessor, Kimi K2.6.

Anastasios Angelopoulos, CEO of Arena.ai, hailed Kimi K3 as "the single biggest release of the year" and a moment where "OSS Chinese models have surpassed US models." This achievement challenges the perception that Chinese AI labs lag behind their American counterparts, especially given that Fable 5 and GPT-5.6 were recently released.

The release of K3 comes amid heightened scrutiny of the U.S.-China AI race and growing national security concerns. It is expected to reignite debates in Washington regarding export controls and the practice of "distillation," where Chinese companies are accused of illicitly extracting capabilities from models like Claude to improve their own.

👉 Read the full article on Gizmodo

Researchers from the Technical University of Denmark, ORCA Computing, and other European institutions have developed a hybrid system where a printer-sized photonic quantum processor modifies a generative AI to design peptides. This innovative approach replaced the usual random noise input for the generative AI with patterns derived from the quantum processor.

The experiment successfully produced short amino acid chains capable of stabilizing specific immune system proteins, a crucial step for developing personalized vaccines and immunotherapies. The most significant improvements were observed in areas where biomedical data is scarce, such as less common HLA variants.

The system, utilizing a Generative Adversarial Network (GAN), used patterns from ORCA Computing's 32-mode photonic quantum processor as a "seed" for the AI's latent space. This quantum-derived input allowed the AI to explore the vast peptide space differently, avoiding repetitive regions and generating more diverse and effective candidates, particularly for HLA alleles with limited training examples.

While the peptides designed by the quantum-enhanced AI showed promising results in laboratory tests, successfully stabilizing peptide-HLA complexes, the researchers emphasize that this does not yet constitute a "quantum advantage." The system was small enough to be simulated classically, and the overall improvements were moderate, though consistent.

This work, published as a preprint on bioRxiv, highlights the potential of quantum computing to enhance AI in data-scarce domains, paving the way for future applications in drug discovery, such as designing synthetic antidotes for snake venoms.

👉 Read the full article on Gizmodo en Español

Google Vids, Google's new AI-powered video creation app, has introduced a feature that allows users to star in their own AI-generated videos. This functionality leverages generative AI to produce personalized video content, making video creation more accessible and engaging for a wide audience.

Users can upload their personal photos or video clips, which the AI then seamlessly integrates into various templates and narrative scenarios. This effectively positions the user as the central figure or protagonist within the AI-created video, offering a unique and personalized storytelling experience.

The tool is designed to streamline video production for both personal and professional applications. It includes a suite of AI-driven features such as automatic script generation, professional-sounding voiceovers, and a range of visual effects, all aimed at simplifying the video creation process.

This enhancement to Google Vids underscores Google's strategic expansion into generative AI applications, particularly those focused on boosting productivity and fostering creativity among its users.

👉 Read the full article on TechCrunch