bibibi.ccLoading…
Connect scattered updates into a story you can understand.
BlockBeats reports that Microsoft has completed pretraining of its self-developed model MAI-Base-2, which uses an MoE architecture and has now entered the post-training stage.
17-year-old AI Researcher Chen Guangyu announces departure from Moonshot AI Kimi team; he previously participated in Attention Residuals、Kimi Delta Attention and K3 related R&D.
Tian Keyu responds 2 responds to the controversy over interfering with ByteDance research projects years ago, saying he only manually interfered 10 times over approximately 5 days; the rebuttal document alleges that he continuously tampered with code and passed off an open-source model as project results.
In validation by Alibaba's Qwen team, QwenGyre Qwen3.8-2.4T, reduced the time for 48 reinforcement learning training steps from 134.55 hours to 75.42 hours, while also improving the model's test pass rate.
Banbury Road released a system consisting of 32 models Kardashev-0.7, saying that joint reinforcement learning enables them to develop different specialties and can reduce inference costs and memory requirements.