PersonalAgentEval launched as a benchmark evaluating 7 personal agents across everyday workflows with realistic user histories. The benchmark measured proactive context retrieval, speed, efficiency, and follow-up compliance.
Key points
Leaders: Instinct and OpenAI Dots topped the leaderboard.
Instinct uncovered 64 of 72 key facts before users mentioned them.
Grok Bot averaged 0.8 minutes of active work per task with a 50% completion rate.
Meta Muse required the fewest average interaction turns.
Sources
- Original post
- Linked resource - Linked in the post
RootSLAM 3–14× faster with Square Root Bundle Adjustment
RootSLAM achieves 3–14× faster performance using Square Root Bundle Adjustment for visual-inertial SLAM. The work appears on arXiv and involves collaborators Tomás Santucci, Prof. Taihú Pire at UNR, and Prof. Daniel Cremers at TUM.
Key points
Accuracy: State-of-the-art SLAM accuracy
Collaboration: Tomás Santucci, Prof. Taihú Pire (UNR), Prof. Daniel Cremers (TUM)
Sources
- Original post
- Linked resource - Linked in the post
VPS prototype with 4K video using AVATA 360 data
VPS (Visual Positioning System) prototype uses 4K video and AVATA 360 data. Map generation was added to 360 MCConverter and a homemade localizer provides self-position estimation. Some parts remain unstable.
Key points
Material: raw material is AVATA 360 data.
Feature: implemented VPS map generation function in 360 MCConverter.
Localization: self-position estimation with a homemade localizer.
Sources
406 researchers open letter calls for AI airgap by default
406 researchers, including Yoshua Bengio, Geoffrey Hinton, Stuart Russell and Ron Rivest, signed an open letter demanding that cyber-capable AI models be airgapped by default until safer internet-access methods are standardized. The letter also addresses models that need outside access.
Key points
406 researchers signed the open letter.
The letter calls for AI models to be airgapped by default until safe internet access is standardized.
