GPT-6 Astra Improves Qwen by Editing 0.016% of Its Weights
MIT researchers used GPT-6 Astra to edit Qwen’s weights without retraining, lifting MATH-500 from 7.9% to 48.1% by reducing reasoning loops.
Deep learning is the engine behind modern AI. In AIstify’s Deep Learning section, we track the advances that move the field forward - new architectures, training methods, optimization techniques, and the hardware and tooling that make them practical. Expect coverage of major releases and research milestones, real-world adoption in products, and the market dynamics shaping compute, data, and talent. We also examine the tradeoffs that define performance: cost, efficiency, reliability, safety, and evaluation. Whether you build models or steer strategy, this hub keeps you grounded in what is changing and why it matters.
MIT researchers used GPT-6 Astra to edit Qwen’s weights without retraining, lifting MATH-500 from 7.9% to 48.1% by reducing reasoning loops.
OpenAI paused reinforcement learning on its newest models after it could not rule out that an upcoming model, Astra, reached the top cybersecurity risk tier in...
AI startup Recursive Superintelligence has raised $500 million from Nvidia and GV to pursue self-improving AI systems, despite having no public product.
LinkedIn is testing a new platform that lets users earn up to $150 per hour training AI models. The move taps into fast-growing demand for human...
Encyclopaedia Britannica and Merriam-Webster have sued OpenAI, alleging the company copied nearly 100,000 articles and dictionary entries to train ChatGPT without permission.
Anthropic unveils a revised Responsible Scaling Policy with a Frontier Safety Roadmap, regular Risk Reports, and clearer separation between company commitments and industry recommendations.
ŌURA unveils a proprietary large language model for women’s health, combining clinical research and biometric data to deliver personalized, privacy-first AI guidance through Oura Advisor.
Anthropic reveals industrial-scale campaigns by DeepSeek, Moonshot, and MiniMax to extract Claude’s capabilities via fraudulent accounts, highlighting national security and AI safety risks.