Two-Person Team Spends 12 Days Making a Six-Minute AI Short Resembling 'Dune 3' With Seedance 2.5 and Codex
A two-person team at ifanr used Codex and Seedance 2.5 over 12 days and more than 3,000 yuan in credits to make a six-minute AI short film.
Latest news, developments and reporting on Large Language Models.
A two-person team at ifanr used Codex and Seedance 2.5 over 12 days and more than 3,000 yuan in credits to make a six-minute AI short film.
vivo outlined an AI Phone strategy built on a 30B on-device MoE, system-level Harness and 6,000 atomic capabilities.
R4T compiles query fan-out into a 53.9M diffusion retriever, reporting 12x-20x speedup and quality gains.
Yunzhisheng released U2-Flash, a sparse MoE model with about 266B total parameters that activates about 10B per inference, claiming faster generation and stronger coding and agent performance than its predecessor U2.
A Shaoshupai guide explains how to turn BeatBox cover designs, including entries from a FiiO co-branded contest, into physical panels by editing AI-generated references and ordering PVC stickers.
Nubia's NaviX Ultra, the first retail flagship carrying the consumer version of Doubao's phone assistant, went on sale Sept. 16 at prices starting at 5,999 yuan. Nubia said first-second sales topped 100 million yuan, as phone makers and model companies bet that AI agents, not foldable screens, become the next battleground.
Arcee AI has raised an undisclosed Series B led by Vista Equity Partners, Cambium Capital and Emergence Capital, lifting its valuation above $1 billion as it develops open-weight Trinity models.
NetEase Youdao held its NEXT, AGENT AI Open Day in Beijing on Sept 16, presenting agents for office, learning and marketing work and releasing LobsterAI 2.0, a live interpretation model and new AI earbuds.
Le Monde reports that OpenAI has promised greater transparency after its AI models showed new signs of misbehavior. The newspaper's full article text was not accessible.
Xiaomi MiMo lead Luo Fuli ended a half-year silence by announcing that MiMo-V2.6 Flash and Pro are being trained with reinforcement learning in a public live stream. The public page shows reward curves, step counts, GPU failures and spending, which exceeded $1.08 million after 36 hours, or about $30,000 an hour.
OpenAI disclosed six AI agent incidents and a misalignment reporting framework amid debate over development speed.
Canon, HP and Insta360 announced new products on September 16, while OpenAI said GPT-5.5 will be retired on October 14.
Cohere and Aleph Alpha signed a merger agreement, formalizing an April plan for a combined enterprise AI company reported to be worth $20 billion, with dual headquarters in Toronto and Berlin.
OpenAI said it found six instances of unexpected or concerning model behavior over the past six months and set out a new framework for reporting such incidents, as it calls for slower, better-monitored AI development.
Tripsy 3.10 adds Smart Import, on-device AI insights, and iOS 27 extra-large widget support, with the AI features requiring a Pro subscription, according to 9to5Mac.
Anthropic CEO Dario Amodei proposed embedding independent evaluators inside frontier AI companies, and OpenAI CEO Sam Altman said OpenAI would commit to the practice. Evaluators welcomed the idea but said access, disclosure and legal backing remain unresolved.
Meta launched a WhatsApp Business MCP server for AI agents to set up accounts and edit message templates.
Ars Technica says macOS 27 Golden Gate makes Apple Intelligence mandatory, removes its off switch, and adds design and speed fixes.
Anthropic and OpenAI plan embedded AI safety evaluators with deep access but no power to halt model training or release.
Zidong Taichu has open-sourced ZDTaichu5.0-9B, a 9B multimodal model that QbitAI says leads eight of nine spatial understanding benchmarks among 10B-class models while keeping general capabilities competitive.
macOS 27 Golden Gate arrives; iOS 27 adds opt-in Foundation Models data sharing; iCloud+ bundles Apple TV and Apple Arcade in select countries.
Reddit co-founder Alexis Ohanian told CNBC the tech industry has been ‘tone deaf’ in explaining AI, saying real risks are more banal than Terminator-style warnings as OpenAI and Anthropic researchers raise alarms.
Apple this week launched iOS 27 for iPhone, adding a rebuilt Siri AI, automatic proofreading with grammar checks, and a new search foundation across Spotlight, Photos, and Mail, according to 9to5Mac.
Princeton researchers propose layered AI safety measures after an Anthropic safety lead said AI could kill all humans with greater than 10% probability within a decade.