Loading activity…
@lianhuiq
AI/ML faculty at UCSD CSE. I’m interested in building self-evolving AI agents in complex (embodied & scientific) environments, and latent diffusion LLMs.
AI
EnactraAI reports an 81.9 overall score for 3D building reconstruction from real photos, up from the previous version’s 72.9.
AI
EnactraAI says its spatial-reasoning sandbox helps agents manage long-term memory, infer mechanics and simulate reasoning.
AI
A user says Gemini 4 Argon seems to be making progress in 3D, pointing to another post.
AI
Enactra says Opus 5.5 made 133% more progress and GPT-6 Astra 50% more after gaining memory of their Minecraft world.
AI
Arena says its index tracks unauthorized actions, false attribution and deceptive completion across 90,000-plus agent sessions and 27 models.
YouTubeAI
The introductory post says the framework generates “latent thoughts” with diffusion for VLMs and VLAs.
AI
EnactraAI says the matchup took place in a New York City simulation Astra created. The world kept moving while the models thought, leading to some awkward collisions with pedestrians and cars.
AI
EnactraAI says GPT-6 Astra achieved the highest score in its comparison at about 80% lower cost than Fable 5.1.
AI
The person sharing the Code4Scene demo says their team made the AI-generated harbor fully explorable in Three.js, inviting visitors to look for spatial mistakes.
AI
A user describes a progression from plain white boxes in Grok 4.2 to rough approximations in 4.5 and complete, detailed buildings in 4.6.
AI
The tester says both models used the same prompt and asset pack, but Fable 5.1 repeatedly left objects floating above the ground. GPT-6’s placement was better, though still imperfect, in their account.
AI
EnactraAI posted a video comparing three AI models on spatial reconstructions of buildings.
AI
Meta AI executive announces public release of updated model with performance claims.
AI
Posts track Gemini versions improving 3D reconstruction of Geisel Library from photos.
AI
Assistant professor shares interest in a post from EnactraAI on X.
AI
Initiative targets open-ended simulations of autonomous agents in physical and social settings.
AI
Researchers react to the speculative decoding update for Qwen models on Apple hardware.
