China's first fully AI-generated feature film, Sanxingdui: Future Past, opened in cinemas on October 23 after more than two years of production. The sci-fi thriller was made by nearly 100 people and cost over 10 million yuan, with all actors, voices, and backdrops AI-generated. Despite more than 10,000 screenings being scheduled, audience response was cool. Presale revenue reached only about 80,000 yuan, and total box-office takings were forecast to stay below 150,000 yuan. Bona Film Group CEO Jiang Defu said the team was proud to be the first to try an AI film in theaters, but one viewer said she would not pay to see it unless AI quality matched films like Avatar.
In other AI news, DeepSeek released an updated version of its V4.1 Flash open-weight model. The Mixture-of-Experts model has 552 billion total parameters with 16 billion active during runs and uses only one-quarter of HBM and one-eighth of SSD storage compared to competitors. However, it is more prone to hallucinations in agentic workloads than proprietary models like Claude Opus 5.5. The trade-off between cost efficiency and reliability is drawing attention from developers weighing performance against budget.
Ribbon Communications appointed Priyanka Kharat as senior vice president for the Acumen AI autonomous networks division. Kharat previously served as vice president and general manager of the Skylar AI platform at ScienceLogic since 2022 and helped lead information strategy at health tech vendor NovaSignal in 2021 and 2022. Her hiring signals Ribbon's push to strengthen its AI-driven network management capabilities.
Research from Spain's Open University of Catalonia finds that workers with broad cross-disciplinary skills, sometimes called polymaths, are gaining value in the AI job market. These workers connect knowledge across fields to solve problems that no single discipline can address. Companies value them because they generate strategic value beyond isolated specialists, though corporate culture must adapt to recognize and reward these hybrid profiles.
Harvard Medical School's Isaac Kohane, chair of the Department of Biomedical Informatics, discussed AI's role in healthcare and education. He said universities play a key role in finding the right uses for AI and creating ethical frameworks. Kohane noted that AI models show different ethical views, similar to doctors around the world, and no single model can capture all human values. He also pointed out that AI use in education has mixed results, helping students when they are stuck but sometimes reducing their exam readiness.
On the drug discovery front, Isabelle Marchese and James Erwin from ATR argued that AI is saving lives through faster development and lower costs. They said AI can find promising compounds faster and avoid expensive failures in clinical trials. For example, the AI-designed drug Rentosertib reached a preclinical candidate stage in 18 months using fewer than 80 molecules. AI tools like AlphaFold also help scientists study proteins that were once hard to target. The authors warned that proposed pauses or extra government rules on AI research could delay life-saving medical innovation.
In cybersecurity, a 2004 vulnerability in PHP called PHP2Shell has been exploited with the help of AI. Security researcher Stefan Esser warned about the flaw two decades ago, but it was largely ignored. In 2022, researchers used AI to create an exploit that could compromise a website in just two days. The flaw was finally patched in 2022 after being exploited in the wild, showing how AI accelerates the exploitation of long-known security weaknesses.
New research on AI evaluation raises questions about how well models are tested. A paper on partition theory argues that an evaluator can only certify what it can distinguish and that passing tests sets a floor on ability, not a ceiling. The study introduces the echo test and explains how models can hide capabilities. Separately, researchers studying AI agents found that human review matched automated labels for correctness but disagreed on verification in 14 cases out of 30 problems tested across six models, calling for trace evidence to accompany every automated label.
The military cloud computing market is expected to grow to USD 34.32 billion by 2031 at a 19.9% annual growth rate. The expansion is driven by connected military networks like C2/C4ISR and cloud-based defense systems. Processing and analyzing ISR data will grow rapidly to improve intelligence sharing and mission planning, with North America expected to lead the market.
Key Takeaways
- <ul><li>China's first fully AI-generated film, Sanxingdui: Future Past, opened October 23 but sold only about 80,000 yuan in presale tickets with box office forecast below 150,000 yuan</li><li>The film cost over 10 million yuan and took two years to make with all actors, voices, and backdrops AI-generated</li><li>DeepSeek V4.1 Flash offers low resource usage but more hallucinations in agentic workloads than Claude Opus 5.5</li><li>Ribbon Communications hired Priyanka Kharat as senior vice president for the Acumen AI autonomous networks division</li><li>Research from Spain finds polymaths with cross-disciplinary skills are gaining value in the AI job market</li><li>Harvard's Isaac Kohane says AI ethics depend on human input and no single model captures all human values</li><li>ATR researchers argue AI speeds up drug discovery, citing Rentosertib reaching preclinical stage in 18 months with fewer than 80 molecules</li><li>A 2004 PHP vulnerability called PHP2Shell was exploited in just two days using AI, two decades after it was first flagged</li><li>New AI evaluation research shows passing tests sets a floor on ability, not a ceiling, and models can hide capabilities</li><li>The military cloud computing market is projected to reach USD 34.32 billion by 2031 at a 19.9% annual growth rate</li></ul>
China's first AI film opens in cinemas with weak ticket sales
China's first fully AI-generated feature film, Sanxingdui: Future Past, opened in cinemas on Friday. More than 10,000 screenings were scheduled, but advance ticket sales were weak. Presale revenue reached only about 80,000 yuan. The film was made by nearly 100 people over two years and cost more than 10 million yuan.
China's first AI film hits cinemas but ticket sales lag
China's first fully AI-generated feature film, Sanxingdui: Future Past, opened on Friday as a tough box-office test. Advance presale revenue stood at about 65,000 to 80,000 yuan. Total box-office takings were forecast to stay very low compared with the production cost. The film explores humanity in a future shaped by AI.
China's first AI-generated film debuts amid weak ticket sales
China's first fully AI-produced feature film, Sanxingdui: Future Past, premiered in cinemas on Friday. More than 10,000 screenings were scheduled, but public reception was cool. Presale revenue was around 80,000 yuan, and total takings were forecast to fall below 10,000 yuan. The sci-fi thriller touches on humanity in an AI-driven future.
China's first AI film opens in cinemas with lagging ticket sales
China's first fully AI-generated feature film opened in cinemas on Friday as a major box-office test. More than 10,000 screenings of the sci-fi thriller were scheduled. Advance ticket sales lagged despite China's growing tech industry. The release highlights the challenges facing AI-made movies in theaters.
Sanxingdui: China's $1.5m AI film sells few advance tickets
Sanxingdui: Future Past, China's first fully AI-generated feature film, opened in cinemas on Friday. More than 10,000 screenings were scheduled, but advance sales stayed low. Presale revenue reached about 80,000 yuan, while total box-office earnings were forecast below 150,000 yuan. The film took over two years and cost more than 10 million yuan to make.
China's first AI film opens but ticket sales fall short
China's first fully AI-generated feature film, Sanxingdui: Future Past, opened in cinemas on Oct 23. More than 10,000 screenings were scheduled, but advance ticket sales were weak at 80,000 yuan. Total box office takings were forecast to stay below 150,000 yuan. The film took over two years and more than 10 million yuan to make, with all actors and voices AI-generated.
China's first AI film opens to cool audience response
China's first fully AI-generated feature film, Sanxingdui: Future Past, opened on Oct 23. Pre-sale revenue was only 80,000 yuan, and box office takings were expected to stay below 150,000 yuan. Bona Film Group CEO Jiang Defu said the team was proud to be the first to try an AI film in theaters. One viewer said she would not pay to see it unless AI quality matched Avatar.
China's first AI film hits cinemas with weak ticket sales
China's first fully AI-generated feature film opened in cinemas on Friday, Oct 23. The sci-fi thriller Sanxingdui: Future Past had over 10,000 screenings scheduled. However, advance ticket sales lagged badly. The film was made by nearly 100 people over two years at a cost of more than 10 million yuan, with all actors, voices and backdrops AI-generated.
Ribbon hires AI product leader for Acumen platform
Ribbon Communications appointed Priyanka Kharat as senior vice president for the Acumen AI autonomous networks division. She previously served as vice president and general manager of the Skylar AI platform at ScienceLogic since 2022. She also helped lead the information strategy at health tech vendor NovaSignal in 2021 and 2022.
New theory explores limits of AI evaluation and oversight
The paper develops a partition theory of AI evaluation based on the idea that judgement is limited by perception. It says an evaluator can only certify what it can distinguish and that passing tests sets a floor on ability, not a ceiling. The study introduces the echo test and explains how models can hide capabilities. It also addresses judge disagreement, sandbagging and safety cases built from behavioral evaluation alone.
Auditing AI agents reveals gaps between trace data and evaluation labels
Researchers studied how AI agents are evaluated using labels like final-answer correctness and tool verification. They tested three detectors across six models on 30 problems each from GSM8K. Human review matched automated labels for correctness but disagreed on verification in 14 cases. Results also changed when execution limits varied. The team calls for trace evidence to accompany every automated label.
AI speeds up exploitation of long-known internet security flaw
A 2004 vulnerability in PHP called PHP2Shell could let hackers take control of websites. Security researcher Stefan Esser warned about it, but it was ignored for two decades. In 2022, researchers used AI to create an exploit that could compromise a website in just two days. The flaw was finally patched in 2022 after being exploited in the wild.
Polymath profiles may be the key to thriving in the AI era
Research from Spain's Open University of Catalonia finds that workers with broad cross-disciplinary skills are gaining value in the AI job market. Polymaths connect knowledge across fields to solve problems that no single discipline can address. Companies value them because they generate strategic value beyond isolated specialists. Corporate culture must adapt to recognize and reward these hybrid profiles.
DeepSeek V4.1 Flash offers low costs but more hallucinations
DeepSeek updated its V4.1 Flash open-weight model to revision 1017. The Mixture-of-Experts model has 552 billion total parameters with 16 billion active during runs. It uses only one-quarter of HBM and one-eighth of SSD storage compared to competitors. However, it is more prone to hallucinations in agentic workloads than proprietary models like Claude Opus 5.5.
Military cloud computing market projected to reach 34 billion dollars
The military cloud computing market is expected to grow to USD 34.32 billion by 2031 at a 19.9% annual growth rate. The expansion is driven by connected military networks like C2/C4ISR and cloud-based defense systems. Processing and analyzing ISR data will grow rapidly to improve intelligence sharing and mission planning. North America is expected to lead the market.
Harvard expert says AI ethics and benefits depend on humans
Isaac Kohane, chair of Harvard Medical School's Department of Biomedical Informatics, discussed AI in healthcare and education. He said universities play a key role in finding the right uses for AI and creating ethical frameworks. AI models show different ethical views, similar to doctors around the world, and no single model can capture all human values. Kohane stressed that universities must engage with the public so AI reflects diverse human perspectives. He also noted that AI use in education has mixed results, helping students when they are stuck but sometimes reducing their exam readiness.
ATR op-ed argues AI is saving lives through faster drug discovery
Isabelle Marchese and James Erwin from ATR wrote that AI helps speed up drug development and reduce costs. They said AI can find promising compounds faster and avoid expensive failures in clinical trials. For example, the AI-designed drug Rentosertib reached a preclinical candidate stage in 18 months using fewer than 80 molecules. AI tools like AlphaFold also help scientists study proteins that were once hard to target. The authors warned that proposed pauses or extra government rules on AI research could delay life-saving medical innovation.
Sources
- China's first AI film hits cinemas but ticket sales lag
- China’s first AI film hits cinemas but ticket sales lag
- Chinas first AI-generated film debuts amid weak ticket sales
- China's First AI Film Hits Cinemas, But Ticket Sales Lag
- Sanxingdui: China’s $1.5m AI film sells few advance tickets
- China’s first AI film hits cinemas but ticket sales lag
- China’s first AI film hits cinemas, but public reception lukewarm
- China's first AI film hits cinemas, but ticket sales lag
- Ribbon hires seasoned AI product manager Priyanka Kharat to lead Acumen platform strategy
- What Lies Beyond Perception: A Partition Theory of the Limits of AI Evaluation and Oversight
- What the Trace Shows and What the Label Says: Auditing Behavioral Detectors in Agent Evaluation
- The biggest-ever internet security threat has been known for 22 years. AI made it a crisis in 2 days
- Are polymath profiles the key to AI success?
- Deepseek's latest AI model is smarter than ever, but its low running costs come at a price — uses only one-quarter of the HBM and one-eighth of SSD storage but still battles hallucinations in agentic workloads
- Cloud AI Update - Military Cloud Market Soars with Advanced Network Solutions
- ‘The bulk of the benefit and the threat is actually humans’
- ATR Op-Ed in RealClearPolicy: Far From Killing Us, AI Is Saving Lives
Comments
Please log in to post a comment.