A research paper by Peking University and YIXIN AI Lab, 'Wrong Prediction, Right Answer: Recovering Evidence from Collapsed LLM Sequence Scores,' was accepted to the main conference of EMNLP 2026 (acceptance rate 15.4%). The study proposes a 'readout bottleneck' diagnostic perspective for LLM reasoning evaluation, showing that models may encode correct answers in hidden states while systematic output-layer biases obscure them, and designs a staged diagnostic method to verify the phenomenon.
At the Beijing Xiangshan Forum on September 17, 2026, former Japanese Prime Minister Yukio Hatoyama voiced opposition to Takaichi Sanae's "AI poisoning," expressing concern that the Japanese government is feeding wrong historical views into the AI business support system "Genai" and warning that AI could become a very dangerous tool if used improperly.
Feng Weide, distinguished research fellow at the Center for Naval Analyses, said at the 13th Beijing Xiangshan Forum that AI is an important dimension of US-China relations, with room for cooperation and competition. He suggested more communication on AI and incorporating AI military applications into US-China military dialogue, calling AI discussion a good example for advancing strategic stability.
At the Silicon Photonics Development Forum of the 19th Pujiang Innovation Forum, Tao Yaoyu from Peking University's Xinyao Yue team unveiled a programmable spatial optoelectronic heterogeneous AI inference chip. It adopts a digital light intensity accumulation and multiply-accumulate split short optical path architecture, targeting low-error, low-loss and programmable AI inference acceleration, achieving first-token latency at the hundred-microsecond level.