New Tsinghua + cornell study shows Agent memory can turn a recoverable mistake into a persistent one. Letting failed agent steps update memory can make the next step inherit the same mistake. This paper borrows a database idea for long-running agents: treat each chunk of work as a transaction, and only commit it after validation. Their ACID-Agent treats each exploration-execution-validation cycle as a transaction, commits only validated results, and keeps failed attempts out of both memory and the workspace. If validation fails, the agent retries without carrying that failed state forward. For long-running agents, reliability may depend less on asking the model to "reason better" and more on controlling what gets committed, what gets retried, and what failed state is allowed to survive. – arxiv. org/abs/2608.13900 Title: "Agentic Transaction: Towards ACID-Compliant Agent Systems"

New Illinois+ Tsinghua University and other labs study finds that LLM agents still have unreliable memory and that it can get worse when they keep rewriting their own memories. LLM agents can learn from experience, but their rewritten memories often become unreliable. The problem is that many agent systems store past work by asking an LLM to compress messy experience into neat written lessons. That sounds useful because the agent should remember what worked before, but the paper finds that repeated rewriting slowly damages the memory. The core idea is that raw episodes, meaning the actual past attempts and solutions, often stay more useful than the polished lessons made from them. The authors tested this across tasks like web shopping, simulated worlds, app use, and ARC-style puzzle problems where they could control the correct solutions. The sharpest result is that GPT-5.4 solved 100% of a small ARC-AGI set with no memory, but after memory was built from correct solutions, streaming updates dropped it to about 54%. The failures came from bad grouping, overbroad lessons, and overfitting, so the memory forgot details, mixed up task types, or learned rules that only worked on narrow examples. The big deal is that agent memory should not automatically rewrite every experience into a summary, because keeping raw evidence and only sometimes making summaries worked better. The paper is really proposing that agent memory should treat raw past episodes as important evidence, not as disposable notes to summarize away. ---- Paper Link – arxiv. org/abs/2605.12978 Paper Title: "Useful Memories Become Faulty When Continuously Updated by LLMs"

"Mooncake originated from a research collaboration between Kimi(Moonshot AI) and Tsinghua University. It was born from the need to solve the 'memory wall' in serving massive-scale models like Kimi K-Series. Since open-sourcing, it has evolved into a thriving community-driven project." GitHub:

International bondholders owed tens of millions of dollars by Beijing-backed semiconductor group Tsinghua Unigroup are seeking to freeze its overseas assets

#Nansha# | 🥮 Sweet taste of Lingnan heritage! 19 international students from Tsinghua, Sun Yat-sen University and the #HongKong# University of Science and Technology (#Guangzhou#) visited Dongchong Town, Nansha, to try their hand at making jianyubing – traditional wedding pastries and local intangible cultural heritage! 👩‍🍳🥟 Guided by master craftsman Liang, they kneaded dough, filled pastries, and watched the baking process up close – gaining a delicious insight into Lingnan water town wedding customs. 💍🌾 #IntangibleCulturalHeritage# #LingnanCulture# #TraditionalFood# #WeddingPastries# #CulturalExchange# #Dongchong# #Tsinghua# #SYSU# #HKUST# #HeritageCraft#

As football fever sweeps the world, a different kind of champion is making headlines. Tsinghua University’s Hephaestus Team successfully defended its Humanoid League title at RoboCup 2026, showcasing China’s progress in humanoid robotics.⚽️🤖