Chinese researchers diagnose AI image models with aphasia-like disorder, develop self-healing framework
Chinese researchers have developed UniCorn, a framework designed to teach multimodal AI models to recognize and fix their own weaknesses.
Read full article about: Chinese AI industry admits US remains ahead for now
Leading figures in China's AI industry are tempering expectations: China won't overtake the US in the AI race anytime soon. Justin Lin, head of Alibaba's Qwen model series, puts the odds of a Chinese company surpassing OpenAI or Anthropic in the next three to five years at less than 20 percent. Tang Jie from Zhipu AI warned at the AGI Next Summit in Beijing that the gap with the US may actually be widening, though recent open-source releases suggest otherwise.
At the conference, executives cited limited computing capacity and US export controls on advanced chips as key hurdles. US infrastructure is one to two orders of magnitude larger, forcing Chinese companies to focus resources on current projects.
Yao Shunyu, a former OpenAI researcher and now Tencent's AI chief scientist, was more optimistic. He cited three to five years as a realistic timeframe for China to catch up but said the lack of advanced chipmaking machines was the main technical hurdle.
The cautious outlook follows a strong week on the stock market. Startups Zhipu AI and MiniMax Group together raised over one billion dollars in Hong Kong, with MiniMax shares doubling on their first day of trading.
Ad
Web world models could give AI agents consistent environments to explore
Researchers at Princeton University, UCLA, and the University of Pennsylvania have developed an approach that gives AI agents persistent worlds to explore. Standard web code defines the rules, while a language model fills these worlds with stories and descriptions.
Read full article about: Convogo's founders join OpenAI to close the gap between AI potential and actual use
OpenAI is bringing in the team behind Convogo, an AI startup that built software for evaluating executives, as part of its broader cloud strategy. Founder Matt Cooper announced the news on LinkedIn. Convogo's software used AI to automatically analyze interviews, surveys, and psychometric tests.
According to OpenAI (via Techcrunch), the acquisition is about the people, not the product. The three founders, Matt Cooper, Evan Cater, and Mike Gillett, will help drive OpenAI's AI cloud efforts. The deal was settled entirely in shares, though the amount remains undisclosed. Convogo's software is being shut down.
The founding team's strong product focus likely made them attractive. Cooper writes that the key to closing the gap between AI's potential and its actual use lies in well-designed, purpose-driven applications, a "usage gap" narrative that Microsoft and OpenAI have both pushed before.
The acquisition also fits OpenAI's strategy of controlling the entire value chain, from infrastructure to models to the end product. This push likely reflects how differentiating on model capabilities alone is getting harder as performance converges and cheaper open-source alternatives catch up.
Comment
Source: via LinkedIn | Techcrunch
Read full article about: OpenAI reportedly sets aside $50 billion for employee stock program
Last fall, OpenAI reportedly set aside a stock pool for employees worth about ten percent of the company. Based on the $500 billion valuation from October 2024, that comes to around $50 billion, according to The Information, citing two people familiar with the plans.
OpenAI has also already issued $80 billion in allocated shares. Combined with the new stock pool, employees now own about 26 percent of the company. Meanwhile, OpenAI is in early talks with investors about a new funding round worth roughly $750 billion.
A previous analysis found that OpenAI pays its employees more than any tech startup in history, with stock-based compensation averaging about $1.5 million per employee. That level of spending complicates the path to profitability: the company is targeting around $20 billion in ARR. But on top of hefty payroll, development costs, and day-to-day operations, OpenAI faces about $1.4 trillion in data center commitments over the next eight years.
Comment
Source: The Information
Ad
New Deepseek technique balances signal flow and learning capacity in large AI models
DeepSeek researchers have developed a technique that makes training large language models more stable. The approach uses mathematical constraints to solve a well-known problem with expanded network architectures.
China captured the global lead in open-weight AI development during 2025, Stanford analysis shows
Chinese open-weight AI is conquering the world: According to a Stanford analysis, models from China have already overtaken their US counterparts in distribution and adoption. But with success come growing geopolitical and security risks.
Ad
AI benchmarks are broken and the industry keeps using them anyway, study finds
Benchmarks are supposed to measure AI model performance objectively. But according to an analysis by Epoch AI, results depend heavily on how the test is run. The research organization identifies numerous variables that are rarely disclosed but significantly affect outcomes.