Revolutionizing Protein Design with PottsMPNN

A new machine-learning framework, PottsMPNN, enhances computational protein design by moving beyond natural sequences to explore a broader range of possibilities.

A new machine-learning framework, PottsMPNN, enhances computational protein design by moving beyond natural sequences to explore a broader range of possibilities.

Multiverse Computing introduces Quantization-Aware Healing (QAH), a novel approach that allows a compressed 4-bit model to outperform its full-precision counterpart, redefining expectations in AI model efficiency.

A new machine-learning algorithm from MIT generates plausible extreme events without relying on historical data, aiding in the preparation for unprecedented scenarios.

Unsloth's new desktop application simplifies the process of downloading, fine-tuning, and exporting AI models, enhancing local AI capabilities without complex setups.

IBM's ALTK-Evolve introduces a nuanced approach to agentic memory, demonstrating that the effectiveness of memory in AI models is not a one-size-fits-all solution.

Recent research highlights the phenomenon of benchmark optimization in automatic speech recognition (ASR) systems, revealing discrepancies between model performance on public benchmarks and real-world applications.

Grok 4.6 enhances long-running agents and interactive tasks, achieving notable benchmarks in coding and knowledge work.

NVIDIA has introduced Cosmos 3 Edge, a groundbreaking model designed to enhance the capabilities of robots and vision AI agents in real-world environments.

MIT researchers unveil GIFT, a system that enhances AI's ability to convert 2D designs into accurate CAD programs, streamlining rapid prototyping.

OpenAI is set to release its latest model series, GPT-5.6, featuring three distinct variants, following government approval.