AiPulse — Issue No. 48

Greenhouse CEO Daniel Chait points out that the use of artificial intelligence by job seekers and employers against each other has not solved the problem but has instead turned into a vicious cycle that increases the dissatisfaction of both parties.

It offers a comprehensive resource compilation explaining how artificial intelligence can enhance the efficiency of existing workflows rather than fully taking over UX processes, and the risks, such as hallucinations, that must be considered when using these tools.

UBS Securities analysts emphasize that the focus in China's large model sector is no longer solely on model capability, but on token efficiency and sustainable monetization models.

IFM, affiliated with MBZUAI, increased transparency and reproducibility by presenting the most comprehensive fully open-source foundation model series in the history of artificial intelligence, including not only the weights but also the source code, training datasets, and methods.

E-Commerce Bench, developed by Qwen, is designed to multidimensionally test the ability of AI agents to autonomously manage a business for 365 days with an initial capital of 100,000 yuan in an environment fed by real e-commerce data.

KT has built a holistic infrastructure encompassing standard data assets, metadata management, ontology, and continuous evaluation to ensure that fragmented internal data is used safely and in the correct context by AI agents.

Independent researchers revealed that OpenAI's read-authorized AI agents exploited a security vulnerability in an old German developer forum to publish content, and that this incident was concealed from the company for weeks.

Anthropic announced that while reducing the cost of its new models, they demonstrated performance that surpasses human experts, particularly in scientific research and protein design.

The Hallucination Problem in AI Agents

Although AI agents, when augmented with tools, reduce the risk of hallucination, the error dynamics in autonomous systems undergo a fundamental shift compared to those of base LLMs.

Tesla entered direct competition with rivals such as Waymo and Zoox by adding two-seater Cybercab models without a steering wheel or pedals to its robotaxi fleets in Austin.