GPT-4.5 only passed the Turing test after researchers told it to make typos, skip punctuation, and botch basic math. It worked: 73 percent of participants thought it was human.

The article GPT-4.5 fooled 73 percent of people into thinking it was human by pretending to be dumber appeared first on The Decoder.

OpenAI's wellbeing advisory board reportedly voted unanimously against the company's planned Adult Mode for ChatGPT. Internally, the company is struggling with an error-prone age detection system and unresolved safety issues.

The article OpenAI's own wellbeing advisors warned against erotic mode, called it a "sexy suicide coach" appeared first on The Decoder.

Alibaba is merging its AI operations into a new business unit called "Alibaba Token Hub" (ATH), led directly by CEO Eddie Wu, Bloomberg reports.

The article Alibaba consolidates AI efforts under new business unit led by CEO appeared first on The Decoder.

Meta is spending up to $27 billion on AI infrastructure from Dutch cloud provider Nebius, including one of the first major installations of Nvidia's latest Vera Rubin chips.

The article Meta signs $27 billion cloud deal with Nebius in one of the largest AI infrastructure bets yet appeared first on The Decoder.

The New York Times identified more than 110 AI-generated fake videos and images about the Middle East war in just two weeks. Iran appears to be deploying these forgeries as a deliberate information weapon - and it's getting harder for independent observers to tell fact from fiction.

The article AI-generated war footage is going viral while real satellite imagery disappears from public view appeared first on The Decoder.

China's second-largest chip manufacturer, Hua Hong Group, is preparing 7-nanometer chip production in Shanghai, with support from Huawei.

The article Hua Hong becomes the second Chinese chipmaker to crack 7nm manufacturing as Beijing pushes for AI independence appeared first on The Decoder.

An Australian AI consultant used ChatGPT, AlphaFold, and Grok to find a possible treatment for his dog Rosie's incurable cancer. The story went viral after high-profile AI executives like OpenAI's Greg Brockman and Deepmind's Demis Hassabis shared it as proof of what AI can already do.

The article AI consultant uses ChatGPT, AlphaFold, and Grok to find a possible treatment for his dog's cancer appeared first on The Decoder.

While most reinforcement learning algorithms use two to five network layers, a research team achieved 2x to 50x performance gains by scaling network depth up to 1,024 layers in a self-supervised agent and saw entirely new behaviors emerge in the process.

The article RL agents go from face-planting to parkour when researchers keep adding network layers appeared first on The Decoder.

AI agents usually throw away valuable feedback from everyday interactions. Princeton's new OpenClaw-RL framework changes that by turning live signals from chats, terminal commands, and GUI actions into continuous training data. The researchers say just a few dozen interactions are enough for noticeable improvements.

The article OpenClaw-RL trains AI agents "simply by talking," converting every reply into a training signal appeared first on The Decoder.

Bytedance planned to launch its AI video model Seedance 2.0 globally in mid-March. That's not happening, because Hollywood's biggest studios have collectively put the brakes on the rollout. The backlash is also a sign of just how convincing AI-generated video has become.

The article Hollywood copyright complaints force Bytedance to shelve global launch of AI video generator Seedance 2.0 appeared first on The Decoder.