x86 formal verification is coming with the first published machine-checked proof of a security property for the end-to-end execution of a Linux process running an x86 binary statically linked against glibc. The work is a joint research project with Zellic.
Key points
First published machine-checked proof of a security property for end-to-end execution of a Linux process running an x86 binary statically linked against glibc.
Joint research project with @zellic_io.
Sources
Recurrent VLMs - prune-then-loop strategy paper
Recurrent VLMs - prune-then-loop strategy paper introduces a method to turn pretrained VLMs into recurrent models. The work is released as an arXiv preprint with a project page.
Key points
Paper: arXiv:2610.22075
Project page announced for the method
Sources
- Original post
- arxiv.org - Linked in the post
- madhavesdios.github.io - Linked in the post
Teachable Figures open-sourced skill for AI agents
Teachable Figures is an open-sourced skill for AI agents to create beautiful, interactive 2D and 3D lessons. It was released for public use.
Key points
Open-sourced Teachable Figures
Skill for AI agents to create interactive 2D and 3D lessons
Sources
ARC3 think-aloud study with 64 participants
ARC3 think-aloud study with 64 participants captured audio of problem solving. Humans and machines solve ARC3 very differently.
Key points
64 participants solved ARC3 while thinking aloud into a microphone.
Humans and machines solve ARC3 very differently.