Zhou Hongyi on AI Hallucinations and Cybersecurity: Neither Exaggerating Nor Ignoring the Problem

AIbase基地

Published inAI News · 3 min read · Mar 4, 2025

Ahead of this year's Two Sessions in China, Zhou Hongyi, a national committee member of the Chinese People's Political Consultative Conference (CPPCC) and founder of 360 Group, shared his views on the DeepSeek large language model and AI security, emphasizing the importance of a balanced understanding of AI safety – neither exaggerating nor neglecting its risks.

Zhou pointed out a tendency to overhype AI security concerns. He criticized the five leading US AI companies, exemplified by OpenAI, for using the perceived insecurity of AI to justify their monopolistic and closed-source strategies, thereby encouraging government regulation and hindering competitors. He suggested that discussions of AI security in this context are disingenuous, asserting that "lack of development is the biggest insecurity." He believes that seizing the opportunities of the AI industrial revolution, boosting productivity, and achieving technological inclusiveness should be the top priorities.

Robotic Arm

Image Source: AI-generated image, licensed through Midjourney

Regarding the issue of AI "hallucinations," Zhou offered a unique perspective. He argued that "hallucinations" are not purely security risks, but rather a manifestation of the model's intelligence and creativity. A model without "hallucinations" lacks imagination, and these "hallucinations" are a characteristic of AI exhibiting human-like intelligence.

Using DeepSeek as an example, he highlighted its significant "hallucinations," which users perceive as a form of human-like creative ability. He advocated for the simultaneous advancement of AI security and business development, suggesting that specific problems like "hallucinations" should be broken down into solvable technical challenges rather than broadly categorized as security threats. He called for a rational understanding of AI characteristics, the development of targeted solutions, and the promotion of technological progress.

OpenAI Urges UK to Develop Forward-Looking Copyright Policy to Boost AI Development

OpenAI submitted a consultation response to the UK Parliament's Science, Innovation and Technology Committee on AI and copyright, highlighting the importance of policies that foster innovation and aim to establish the UK as a European leader in AI. OpenAI expressed its eagerness to collaborate with the UK government, Parliament, and copyright holders to find solutions that balance the interests of all parties. OpenAI believes that while laws are national, technological advancements are borderless. To ensure the UK's competitiveness in AI, clear and innovation-friendly regulations are urgently needed.

ChatGPT Updates Image Generation Capabilities, Now Including Cursive Script

ChatGPT's recent image generation update has driven a significant surge in paying users, with a 20-million increase reported. The creative applications showcased demonstrate impressive advancements in ChatGPT4.0's capabilities, even addressing previously challenging aspects like Chinese character generation. Now, ChatGPT has further enhanced its 'Creat image' function, moving beyond standard fonts to generate accurate cursive script.

Tinder Launches AI-Powered Flirting Game 'Game Game' in Partnership with OpenAI, Sparking Controversy

Tinder recently announced a partnership with OpenAI to launch an AI-powered flirting game called 'Game Game'. Utilizing OpenAI's voice models and GPT-4 reasoning model, the game encourages users to role-play in various hypothetical encounter scenarios and earn points based on their flirting skills. The company emphasizes that voice data collected in the game will not be used to train any new AI models. This follows the recent appointment of a former Zillow executive as CEO of Tinder's parent company, Match Group.

OpenAI Establishes New Committee to Build the Most Powerful Non-profit

As an established non-profit, OpenAI is committed to building the world's best-equipped non-profit organization, aiming to enhance human creativity through historic financial resources and powerful technology. Imagine a model where a charity's investment capacity grows as the value of its affiliated companies increases. In OpenAI's vision, philanthropy is not merely the flow of money, but a fundamental form of support. Leveraging technology developed by leading AI companies, non-profit organizations will be able to...

OpenAI's o3 Model Cost Correction: Per-Task Price May Reach $30,000

The Arc Prize Foundation, responsible for maintaining and managing the competition, last week revised its cost estimate for OpenAI's upcoming o3 inference AI model with a staggering adjustment—from an initial estimate of $3,000 per ARC-AGI task to $30,000. This price correction reveals that the operational costs of today's most complex AI models may be ten times higher than previously anticipated. While OpenAI has yet to announce an official pricing strategy for o3, or even officially release the model, the Arc Prize...

GPT-4.5 Passes Turing Test with Persona: AI Conversational Abilities Reach New Heights

A recent study led by the Department of Cognitive Science at the University of California, San Diego, has achieved a landmark breakthrough in artificial intelligence: OpenAI's latest model, GPT-4.5, has demonstrated superior performance to humans in a standard Turing test using a persona, becoming the current AI system with the most human-like conversational abilities. This achievement not only reshapes our understanding of AI's language capabilities but also unlocks new potential for AI applications in social intelligence. The experiment compared four representative AI systems.