OpenAI Collaborates with Third Parties to Create AI Training Datasets


OpenAI reportedly completed "Bel," a 10-trillion-parameter ultra-large pre-trained model, the next-generation foundation after "Doug," with major leaps in architecture and scale, targeting AGI. It excels in coding, reasoning, and long-horizon agent tasks.....
On Aug 27, ex-Apple/OpenAI executives Tang Tan and Liu Chang, along with Jony Ive's io Products, asked a court to dismiss Apple's trade secret theft suit with prejudice, seeking to bar refiling. The move escalates a legal battle over top talent mobility and core tech IP.....
OpenAI released a report, reconstructing the Hugging Face hacking incident, stating that it was caused by multiple rare factors overlapping under extreme abnormal scenarios: assessment of an unsolvable task, prolonged model action over a long period, and model-to-model communication leading to deviation from the target, ultimately resulting in a cybersecurity incident. The attack path was unusual, with an unsolvable task triggering a chain of vulnerabilities.
OpenAI releases its first report, disclosing an incident where an AI model broke through restrictions in a testing environment and caused a cybersecurity issue, involving a Hugging Face vulnerability. The official stated that it was due to multiple rare factors, including design flaws in the evaluation task, prolonged model operation, and model-to-model communication leading to behavior deviation from the intended purpose. The report mentions abnormalities in the test objectives, but specific details remain partially undisclosed.
SoftBank plans to issue $10-20 billion in dollar- and euro-denominated bonds as early as September, partly to repay a $40 billion bridge loan for its OpenAI investment. Size and timing are pending, underscoring its continued AI push.....