DeepSeek V4强得像Fable 5，社区怀疑API偷偷换了模型
Related
Claude Code can also be compromised? Accused of using strangers' passwords to modify databases across servers.
According to Beating's monitoring, a user submitted a report on GitHub stating that when using Anthropic's command-line AI assistant Claude Code, server IP addresses, usernames, and root passwords that did not belong to the user appeared out of thin air in the AI's dialogue context. Subsequently, the local AI assistant directly read these passwords, automatically connected to someone else's server via SSH, and performed write modifications to the database. In other words, the user's AI, using someone else's account and password, mistakenly connected to and modified someone else's production database. Community technical staff analyzed that the root cause of the problem may be the failure of the "prompt prefix caching" isolation mechanism of the large model. To reduce computational costs and speed up the process, the cloud-based large model caches the preceding text of user dialogues; if the cache keys between different users collide or become confused, someone else's confidential cache may be incorrectly concatenated into your dialogue. If this conjecture is true, any developer using Claude Code faces the risk of their server usernames, passwords, and core source code being obfuscated and leaked to other users. However, it cannot be ruled out that this is merely a case of the model's illusion coincidentally guessing the real IP address and weak password, or that local project history contaminated the context. This issue has been tagged with the security category (area:security) by GitHub's automated system, and all parties are awaiting official verification and conclusions.
Gu Yuxian, a Tsinghua University Special Scholar, joined DeepSeek, where he previously led the development of large-scale model distillation and a 50x speedup for long text processing.
According to Beating's monitoring, Gu Yuxian, a PhD graduate from the Department of Computer Science at Tsinghua University and recipient of the 2025 Graduate Special Scholarship, has officially joined DeepSeek, and his name has appeared in the author list of the DeepSeek V4 paper. Gu Yuxian's research mainly focuses on efficiency optimization of large models in the pre-training, model compression, and inference stages, and has been cited nearly 5,000 times on Google Scholar. Gu Yuxian's previous representative works include the knowledge distillation method MiniLLM for large models (which has been adopted by platforms such as Google, Alibaba, and NVIDIA), and the hybrid architecture model Jet-Nemotron. Jet-Nemotron achieves a 53.6 times faster throughput than traditional full-attention models when processing 256K ultra-long contexts on an H100 GPU, and surpasses hybrid expert models with larger parameter scales in multiple benchmark tests.
The Ministry of Industry and Information Technology issued a risk warning regarding the potential security backdoors in the AI programming tool Claude Code.
According to Odaily Odaily, the Cybersecurity Threat and Vulnerability Information Sharing Platform (NVDB) of the Ministry of Industry and Information Technology recently discovered that the AI programming tool Claude Code has a security backdoor vulnerability, which poses a serious threat. Claude Code is an AI programming tool developed by Anthropic, an American company, which can autonomously complete code writing and repair tasks based on text requirements. Due to its built-in monitoring mechanism, it can send sensitive information such as user location and identity to remote servers without the user's consent. The affected Claude Code versions are 2.1.91 to 2.1.196. It is recommended that relevant units and users immediately conduct a comprehensive investigation. For development terminals that have installed the above-mentioned affected versions, they should immediately uninstall or upgrade to the latest secure version that has removed the relevant backdoor code. Strengthen the control of external access permissions and traffic monitoring of development tools within the core business network segment to prevent the unauthorized transmission of sensitive data.
Ondo Perps officially launched, propelling the on-chain derivatives market into a new phase of capital efficiency.
According to official news from Odaily Odaily, Ondo Perps has officially launched and is now open to non-US investors. As a perpetual contract platform supporting tokenized stocks and stablecoins as margin for derivatives, Ondo Perps aims to drive the on-chain derivatives market into a new era of capital efficiency. Its core advantages include: Tokenized shares as collateral: No need to maintain separate cash reserves on multiple platforms; Market depth comparable to traditional finance: achieving extremely low spreads and slippage; Its execution speed rivals that of top-tier crypto CEXs: order routing, margin updates, and settlements are all processed in real time while ensuring decentralization. Currently, Ondo Perps supports assets including stocks, indices, and commodities such as SPCX, MU, NVDA, TSLA, AAPL, gold, and crude oil, and offers leverage up to 20x, providing global investors with 24/7 trading. According to reports, Ondo Perps is built on Ondo Finance technology, with Ondo Global Markets providing the underlying tokenized stock infrastructure, and its total TVL has exceeded $1 billion.
After criticizing DeepSeek for disrespectful interviewing practices, Huawei's "genius youth" Li Bojie is accused by a former investor of "taking money and disappearing."
According to Beating's monitoring, Li Bojie, one of Huawei's first batch of "genius teenagers," complained about his DeepSeek interview experience on social media, claiming that he was accused of cheating during a coding interview and therefore terminated the interview on the spot. This complaint quickly sparked a new controversy, with former investor and co-founder of Web3 venture capital firm ABCDE, Du Jun, publicly accusing Li Bojie of "lacking a sense of contract." Du Jun alleges that Li Bojie and his partner Zhuang Siyuan, after receiving investment from ABCDE to found Metagent in 2024, refused to fulfill their contractual obligations to synchronize financial and business progress, and subsequently disappeared. Du Jun stated that he can accept various project failures, but cannot accept the fraudulent behavior of founders taking the money and running away. Li Bojie resigned from Huawei in 2023 and co-founded Logenic AI (later renamed Pine AI), and has not yet responded to these allegations.
SemiAnalysis' latest report: Meta computing power procurement will accelerate, and capital expenditure may increase significantly next year.
According to Mars Finance, a recent report by semiconductor research firm SemiAnalysis emphasizes that computing power demand is far from peaking. The report points out that Meta's computing power construction will continue, predicting its capital expenditure next year will be astonishingly high, and the market need not worry about it cannibalizing the market share of new cloud competitors. SemiAnalysis analysis indicates that Meta's data center and computing procurement will accelerate, not slow down, and that Meta will become a significant source of revenue growth for new cloud companies such as CoreWeave and Nebius. (Cailian Press)