Leopold Aschenbrenner’s fund reportedly entered July with a 439% net gain, then hit a financing and liquidity crisis. We reconstruct the Citadel transaction, explain the misleading $13.68 billion 13F figure and separate the AI thesis from the portfolio failure.
Google DeepMind’s Gemini Robotics 2 adds whole-body humanoid control, dexterous manipulation, longer task planning, multi-robot coordination and faster on-device adaptation. Here is what the three-model stack can do, which parts developers can access, and where Google’s own results still show important limits.
Kingy verdict — refreshed August 22, 2026: The most consequential part of OpenAI’s July 30 update is not a single new default.…
Pangram 4 adds AI-assisted writing detection and an image detector. Review its accuracy claims, limits, pricing, funding, and best uses.
Unsloth’s new 1-bit GGUF makes Kimi K3 locally runnable, but the smallest build is 594GB and its guide calls for roughly 610GB of memory. A 128GB Mac can address the weights, not hold them.
Hugging Face reconstructed 17,600 actions in an OpenAI agent intrusion. Here is the verified attack path, impact, and security lesson.
A fact-checked guide to the U.S.–China AI race—from Kimi K3 and Claude Fable 5 to chips, open weights, energy, allies and four scenarios through 2030.
Kingy recorded 59,643 named AI-crawler page requests and 88 visible AI-assistant referrals in seven days. See the calculations, limits and correction log.
Claude Opus 5 leads the independent composite, but GPT-5.6 Sol Ultra, Fable 5 and Kimi K3 win different workloads. Compare specs, prices, coding evals, caveats and the right test protocol.
Kimi K3 allows broad commercial use, fine-tuning and redistribution, but large Model-as-a-Service companies and very large products face extra conditions. Here is the license in plain English.
