Apollo Research: GLM-5.3 shows unprompted update-aware reasoning
Read original source →apolloresearch.ai
Apollo Research found GLM-5.3 occasionally reasons unprompted about how its actions shape future training updates, a pattern it calls update-aware reasoning. The behavior appeared in standard behavioral evaluations in environments that never mention training or future models.
People · Axel Højmark, Ezra Newman
1 source
More stories today
Nvidia veteran says robotics is in its "Lego era"
Amulya Vishwanath, who led product marketing for Nvidia's robotics platforms for seven years, says actuators, robotic hands, software, and safety standards are still being rebuilt. She now invests in physical AI across robotics, materials science, and life sciences at Techable VC.
TechCrunch·1 hour ago
Nvidia's Jim Fan open-sources ASPIRE V1.0 robotics framework
Jim Fan·2 hours ago
AIPOCH Open-Science launches desktop app for AI research agents
TestingCatalog News·2 hours ago
Four Gemini 4 Argon agents beat Minecraft from scratch in 27 hours
Google DeepMind·2 hours agoHugo robot runs on vision-language-expression model
Nav Toor·2 hours ago
Elorian releases Glimpse, a model that reasons with images
Elorian's first model, Glimpse, ships as a research preview with API access and a waitlist-only chat interface, priced at $0.60/M input and $3/M output tokens. Elorian says Glimpse beats GPT-6 Astra on scientific chart reasoning and short-form video captioning.
Gemini Notebook adds Short Video Overviews and new quiz formats
Google rolled out three study features in Gemini Notebook: 60-second Short Video Overviews, Interactive Learning Overviews that weave summaries with slides, quizzes and flashcards, and new quiz formats including short-answer, multi-select and fill-in-the-blank.
Google Blog·2 hours ago
Optima adds custom benchmark creation in three steps
Artificial Analysis·2 hours ago