
快手可灵也吃上了香蕉,一通离谱prompt测试,好好玩要爆了 (opens in a new tab)
ChatGPT发布三周年,OpenAI没发布,各大AI玩家倒纷纷整出大活。
Total 1186 articles

ChatGPT发布三周年,OpenAI没发布,各大AI玩家倒纷纷整出大活。

论文提出的方法名为 RL4HS,它使用了片段级奖励(span-level rewards)和类别感知的 GRPO(Class-Aware Group Relative Policy Optimization),从而避免模型偷懒、只输出无错误预测。

Vals AI reports that GPT-6.2 Astra formalized Smoktunowicz’s established theorem in 1,794 lines of Lean during Autoform-Hard. Part of its argument differs from the 2006 proof, but the route’s relationship to prior literature still needs further checking.

OpenAI introduces Astra for complex professional work and Sol for everyday tasks, with standard API prices and benchmark results. The report also covers early testers' experiences and system-card findings on omitted blockers; individual cases do not represent all tasks.

Sierra and Meta's personal agent protocol addresses access, while OpenAI's checkout flow requires user payment confirmation. Public information has not established whether the two systems interoperate.

A Shanghai Jiao Tong University team compares its roughly 13.2M-parameter MetaPred language model with a same-size GPT baseline. The work offers initial engineering evidence, not validation of a complete AGI system.

The study evaluates diagnostic and treatment strategies for 40 unpublished complex cases. Scores improved when 250 doctors used the system, but the results do not establish real treatment outcomes.

LongChang 构建 15 个 MuJoCo 场景,其中 14 个各改写一条物理规律。在 12 个被测模型中,仅 GPT-6 Astra 得出了有实验支持的规律发现,发现率为 16.7%。

AISLE 团队使用 4,079 项测试评估智能体操作真实实验设备的能力。GPT-5.5 在该框架中得分 41.19,所测模型未能完全通过安全检查。

网友用空房间图片测试个性化生成。研究人员也曾用捐赠对话样本分析保存的记忆条目,样本中的比例和当前产品行为需分别理解。