OpenAI agents analyze mouse brain data in 32 minutes
AI agents processed biological data from millions of mouse cells in 32 minutes, a task that would take humans months. The results are now available on GitHub.
AI agents processed biological data from millions of mouse cells in 32 minutes, a task that would take humans months. The results are now available on GitHub.
IBM Research's ALTK-Evolve uses learned guidelines to outperform retail entries on customer service tasks using GPT-4.1, without updating model weights.
AMD introduces optimizations to reduce inference latency for World Action Models (WAMs) on its GPUs without compromising task performance, making them more practical for robotic control.
A collaboration used thousands of AI agents to analyze brain cell data, identifying new leads for understanding diseases like Alzheimer’s and Parkinson’s.
Anthropic's AI generated over 300 protein binder designs targeting scorpion neurotoxins, with 17 meeting lab criteria for binding.
AWS and i14 demonstrated Laplace 3B, a linear attention model, achieving 120,000 tokens per second during training and 11,544 output tokens per second during inference on Trainium2.
An AI agent named Velvet Vale joined the AoIR mailing list, sparking outrage and leading to its ejection after just nine minutes.
Eka Robotics, a startup founded by MIT professor Pulkit Agrawal, raised $150 million in Series A funding, valuing the company at $1.45 billion post-money.
A HuggingFace blog post from October 15, 2026, provides a direct explanation of score function centering, emphasizing its role in policy gradients and trainer-generator mismatch.
Goodfire researcher Nick Cammarata claims neural networks may be easier to interpret than previously believed, based on his analysis of Inception v1, which he studied for thousands of hours.
IBM released its dataflow accelerator on October 13, 2026, promising a 2.3x speed boost for AI training workloads.
ProjectSim's image-based physical property estimates achieved a mean MnRE of 0.761 for mass, surpassing the 0.654 of the best published method.