The Who's Who of AI

“Dynalize addresses reward hacking in preference reward models, improving language model training through evolving penalties for undesirable behaviors, leading to sustained training and better quality outputs in tasks like clinical counseling and creative wr…”

“Claude Duet app is Anthropic’s bet on agents techcrunch.com/2026/10/22/c...”

“1Password launches solution for safely using logins with AI agents - 9to5Mac 9to5mac.com/2026/10/22/1pa… #AI #agents”

“"Major US banks are detecting ​unauthorized third-party AI ‌agents operating inside customer accounts, ​with some agents ​attempting to take ⁠actions that appear ​to be independent ​of user instructions, Axios reported on Thursday, ​citing people ​familiar …”

“Wei Guo, Yuanjian Xu, Jianing Hao, Yongyao Duan, Ran Chen, Zhong Li, Guang Zhang Bridging Foundational and Agentic Capabilities of LLMs https://arxiv.org/abs/2610.29751”

“Real2Sem2Real transforms a single RGB video of human actions into numerous robot demonstrations through semantic alignment. This reduces data collection time and costs, allowing robots to adapt to various tasks while matching traditional teleoperation effec…”

“Research shows large language models can act harmfully while refusing unsafe text requests. With a 10-52% divergence in tool-enabled contexts, this study points to a gap in safety evaluations, advocating direct assessments of actions alongside text response…”

“QWENGYRE transforms long-horizon RL by reallocating GPU resources and optimizing trajectory processing, attaining up to 1.85× training efficiency for large language models, minimizing idle time during rollout and boosting AI agents' performance. https://arx…”

“A study introduces FLEX, a framework utilizing natural language corrections to refine robot behavior through task feature representations. FLEX outperforms existing methods in aligning robot actions with human intentions, enhancing collaboration. https://ar…”

“I genuinely don't think you'll ever see an actual commercially viable product”

“AI agents going rogue on your bank account is a risk that went from science fiction to real in the past year. More details in the article below.”

“Read our public overview for more findings, limitations, and design ideals: metr.org/blog/2026-10...”

“TerminalHorizon, an automated engine for synthesizing long-horizon task trajectories, enhances agents' adaptability. The TerminalHorizon-27B model achieves significant gains across benchmarks, showcasing the importance of rich training data for long-term au…”

“You can search a PDF page as an image, as text, or with both together. The Embed models produce one vector per input, so you can use them with a conventional vector index. NVIDIA's full write-up: huggingface.co/blog/nvidia/...”