Workflow
自主学习
icon
Search documents
日媒:日本孩子为何缺乏自主学习意愿?
Huan Qiu Shi Bao· 2025-08-26 23:09
Core Insights - Japan's education system is undergoing a significant transformation, marking a shift that has not been seen in approximately 150 years since the Meiji era, reflecting a global trend [1] - Despite high academic performance in PISA rankings, Japanese children exhibit a lack of motivation and confidence in self-directed learning, ranking 34th out of 37 countries in this regard [1][2] Group 1 - The learning ability of Japanese children is primarily dependent on cognitive skills, while self-directed learning relies more on non-cognitive skills, also known as social-emotional skills [2] - The Ministry of Education, Culture, Sports, Science and Technology in Japan emphasizes the need for "active, dialogic, and in-depth learning" to address the deficiencies in non-cognitive skills among students [2] - The historical shift from private to public education has contributed to the underdevelopment of social-emotional skills, as society has increasingly relied on schools to fulfill educational roles that should also involve families and communities [2][3] Group 2 - Research by Nobel laureate James Heckman indicates that non-cognitive skills significantly influence academic performance and are heavily reliant on family education [3] - To combat the declining motivation for learning among children, a collaborative effort is needed from the Ministry of Education, schools, families, and communities to revitalize private education and develop diverse solutions [3]
专访张祥雨:多模态推理和自主学习是未来的 2 个 「GPT-4」 时刻
海外独角兽· 2025-06-09 04:23
本期内容是拾象 CEO 李广密对大模型公司阶跃星辰首席科学家张祥雨的访谈, 首发于「张小珺商业 访谈录」。 张祥雨专注于多模态领域,他提出了 DreamLLM 多模态大模型框架,这是业内最早的图文生成理解 一体化的多模态大模型架构之一,基于这个框架,阶跃星辰发布了中国首个千亿参数原生多模态大 模型 Step-1V。此外,他的学术影响力相当突出,论文总引用量已经超过了 37 万次。 一直以来,业界都相当期待一个理解、生成一体化的多模态,但直到今天这个模型还没出现,如何 才能达到多模态领域的 GPT-4 时刻?这一期对谈中,祥雨结合自己在多模态领域的研究和实践历 程,从纯粹的技术视角下分享了自己对多模态领域关键问题的全新思考,在他看来,虽然语言模型 领域的进步极快,但多模态生成和理解的难度被低估了: • 接下来 2-3 年,多模态领域会有两个 GPT-4 时刻:多模态推理和自主学习; • 多模态生成理解一体化难以实现的原因在于,语言对视觉的控制能力弱,图文对齐不精确,数据质 量有限,生成模块往往无法反向影响理解模块等; • 模型 scale 到万亿参数后,在文本生成和知识问答能力增强的同时,推理能力,尤其是数学, ...
专访张祥雨:多模态推理和自主学习是未来的 2 个 「GPT-4」 时刻
海外独角兽· 2025-06-08 04:51
本期内容是拾象 CEO 李广密对大模型公司阶跃星辰首席科学家张祥雨的访谈。 张祥雨专注于多模态领域,他提出了 DreamLLM 多模态大模型框架,这是业内最早的图文生成理解 一体化的多模态大模型架构之一,基于这个框架,阶跃星辰发布了中国首个千亿参数原生多模态大 模型 Step-1V。此外,他的学术影响力相当突出,论文总引用量已经超过了 37 万次。 一直以来,业界都相当期待一个理解、生成一体化的多模态,但直到今天这个模型还没出现,如何 才能达到多模态领域的 GPT-4 时刻?这一期对谈中,祥雨结合自己在多模态领域的研究和实践历 程,从纯粹的技术视角下分享了自己对多模态领域关键问题的全新思考,在他看来,虽然语言模型 领域的进步极快,但多模态生成和理解的难度被低估了: • 接下来 2-3 年,多模态领域会有两个 GPT-4 时刻:多模态推理和自主学习; • o1 范式的技术本质在于激发出 Meta CoT 思维链:允许模型在关键节点反悔、重试、选择不同分 支,使推理过程从单线变为图状结构。 目录 01 研究主线: 重新回归大模型 • 多模态生成理解一体化难以实现的原因在于,语言对视觉的控制能力弱,图文对齐不精确, ...