大模型 2.0:展望Gemini 2/GPT Next, 從Token到task-oriented intelligence
(以下内容从天风国际证券《大模型 2.0:展望Gemini 2/GPT Next, 從Token到task-oriented intelligence》研报附件原文摘录)
Next-level large language models 大模型 2.0 : 展望Gemini 2/GPT Next, 从Token到task-oriented intelligence Up ahead: Gemini 2.0 vs GPT; from token to task-oriented intelligence 投资要点/Investment Thesis 投资要点/Investment Thesis 我们认爲从Intelligence Per Token到Inlligence Per Task的模型变化是重要方向。 1)训练阶段,大模型训练预计继续遵循“Scaling Law”。云、创业公司、主权AI将继续大量采购算力并建设数据中心,追求继续提升Intelligence Per Token,预计全球训练集羣将持续扩容,在训练集羣超过10万片GPU後,对数据中心的建造、部署、供电、互联构成新挑战。计算密度、互联密度、功率散热密度与存储密度将有望快速提升。 We believe the step change in computing from token to task-oriented performance represents an important development direction. Training: during the training stage, we expect large language models (LLM) will heed the scaling law. In this regard, cloud, startup and sovereign AI businesses would continue to purchase large amounts of computing power and build data centers, striving to increase computing intelligence with heavier token consumption. Training clusters around the world would keep growing larger, but once they surpass 100,000 GPUs in size, we believe new challenges will arise for data center construction, deployment, power supply and interconnectivity, as density requirements increase greatly for computing performance, interconnectivity, power heat dissipation and storage. 2)推理阶段,我们认爲树状搜索/自博弈等方式有望快速提升Intelligence Per Token,我们看好在计算阶段GPU与CPU的高速互联。以及在应用阶段,我们认爲模型新架构带来的规划能力逐步提高;数学、代码与通用能力也将逐步提升。 Inference: we believe tree-based search in self-play and other methods will improve rapidly with token-driven intelligence: ·At the computation stage, we are bullish about high-speed interconnectivity between GPUs and CPUs. ·At the application stage, we believe planning capabilities of new model architectures will improve over time, accompanied by improvements in math, coding and general skills. 投资建议/Investment Ideas 我们建议关注标的:NVDA MSFT GOOG AMD ARM CFLT ESTC Stock ideas: we would monitor NVIDIA (NVDA US, BUY) and Microsoft (MSFT US, BUY). Not-rated market players: Google (GOOGL US), Advanced Micro Devices (AMD US), Arm Holdings (ARM US), Confluent (CFLT US), Elastic NV (ESTC US). 风险提示:行业竞争加剧;AI发展与需求不及预期;产品交付不及预期 Risks include: intensifying competition; worse AI development progress or weaker demand than expected; and product deliveries falling short of expectations. Email: equity@tfisec.com TFI research report website: (pls scan the QR code) 本文件由天风国际证券集团有限公司, 天风国际证券与期货有限公司(证监会中央编号:BAV573)及天风国际资产管理有限公司(证监会中央编号:ASF056)(合称“天风国际集团”)编制,所载资料可能以若干假设为基础,仅供作非商业用途及参考之用途,会因经济、市场及其他情况而随时更改而毋须另行通知。任何媒体、网站或个人未经授权不得转载、连结、转贴或以其他方式复制发表本档及任何内容。已获授权者,在使用本档或任何内容时必须注明稿件来源於天风国际集团,并承诺遵守相关法例及一切使用的国际惯例,不为任何非法目的或以任何非法方式使用本档,违者将依法追究相关法律责任。本档所引用之资料或资料可能得自协力厂商,天风国际集团将尽可能确认资料来源之可靠性,但天风国际集团并不对协力厂商所提供资料或资料之准确性负责。且天风国际集团不会就本档所载任何资料、预测及/或意见的公平性、准确性、时限性、完整性或正确性,以及任何该等预测及/或意见所依据的基准作出任何明文或暗示的保证、陈述、担保或承诺而负责或承担任何法律责任。本档中如有类似前瞻性陈述之内容,此等内容或陈述不得视为对任何将来表现之保证,且应注意实际情况或发展可能与该等陈述有重大落差。本档并非及不应被视为邀约、招揽、邀请、建议买卖任何投资产品或投资决策之依据,亦不应被诠释为专业意见。阅览本文件的人士或在作出任何投资决策前,应完全瞭解其风险以及有关法律、赋税及会计的特点及後果,并根据个人的情况决定投资是否切合个人的投资目标,以及能否承担有关风险,必要时应寻求适当的专业意见。投资涉及风险。敬请投资者注意,证券及投资的价值可升亦可跌,过往的表现不一定可以预示日後的表现。在若干国家,传阅及分派本档的方式可能受法律或规例所限制。获取本档的人士须知悉及遵守该等限制。
Next-level large language models 大模型 2.0 : 展望Gemini 2/GPT Next, 从Token到task-oriented intelligence Up ahead: Gemini 2.0 vs GPT; from token to task-oriented intelligence 投资要点/Investment Thesis 投资要点/Investment Thesis 我们认爲从Intelligence Per Token到Inlligence Per Task的模型变化是重要方向。 1)训练阶段,大模型训练预计继续遵循“Scaling Law”。云、创业公司、主权AI将继续大量采购算力并建设数据中心,追求继续提升Intelligence Per Token,预计全球训练集羣将持续扩容,在训练集羣超过10万片GPU後,对数据中心的建造、部署、供电、互联构成新挑战。计算密度、互联密度、功率散热密度与存储密度将有望快速提升。 We believe the step change in computing from token to task-oriented performance represents an important development direction. Training: during the training stage, we expect large language models (LLM) will heed the scaling law. In this regard, cloud, startup and sovereign AI businesses would continue to purchase large amounts of computing power and build data centers, striving to increase computing intelligence with heavier token consumption. Training clusters around the world would keep growing larger, but once they surpass 100,000 GPUs in size, we believe new challenges will arise for data center construction, deployment, power supply and interconnectivity, as density requirements increase greatly for computing performance, interconnectivity, power heat dissipation and storage. 2)推理阶段,我们认爲树状搜索/自博弈等方式有望快速提升Intelligence Per Token,我们看好在计算阶段GPU与CPU的高速互联。以及在应用阶段,我们认爲模型新架构带来的规划能力逐步提高;数学、代码与通用能力也将逐步提升。 Inference: we believe tree-based search in self-play and other methods will improve rapidly with token-driven intelligence: ·At the computation stage, we are bullish about high-speed interconnectivity between GPUs and CPUs. ·At the application stage, we believe planning capabilities of new model architectures will improve over time, accompanied by improvements in math, coding and general skills. 投资建议/Investment Ideas 我们建议关注标的:NVDA MSFT GOOG AMD ARM CFLT ESTC Stock ideas: we would monitor NVIDIA (NVDA US, BUY) and Microsoft (MSFT US, BUY). Not-rated market players: Google (GOOGL US), Advanced Micro Devices (AMD US), Arm Holdings (ARM US), Confluent (CFLT US), Elastic NV (ESTC US). 风险提示:行业竞争加剧;AI发展与需求不及预期;产品交付不及预期 Risks include: intensifying competition; worse AI development progress or weaker demand than expected; and product deliveries falling short of expectations. Email: equity@tfisec.com TFI research report website: (pls scan the QR code) 本文件由天风国际证券集团有限公司, 天风国际证券与期货有限公司(证监会中央编号:BAV573)及天风国际资产管理有限公司(证监会中央编号:ASF056)(合称“天风国际集团”)编制,所载资料可能以若干假设为基础,仅供作非商业用途及参考之用途,会因经济、市场及其他情况而随时更改而毋须另行通知。任何媒体、网站或个人未经授权不得转载、连结、转贴或以其他方式复制发表本档及任何内容。已获授权者,在使用本档或任何内容时必须注明稿件来源於天风国际集团,并承诺遵守相关法例及一切使用的国际惯例,不为任何非法目的或以任何非法方式使用本档,违者将依法追究相关法律责任。本档所引用之资料或资料可能得自协力厂商,天风国际集团将尽可能确认资料来源之可靠性,但天风国际集团并不对协力厂商所提供资料或资料之准确性负责。且天风国际集团不会就本档所载任何资料、预测及/或意见的公平性、准确性、时限性、完整性或正确性,以及任何该等预测及/或意见所依据的基准作出任何明文或暗示的保证、陈述、担保或承诺而负责或承担任何法律责任。本档中如有类似前瞻性陈述之内容,此等内容或陈述不得视为对任何将来表现之保证,且应注意实际情况或发展可能与该等陈述有重大落差。本档并非及不应被视为邀约、招揽、邀请、建议买卖任何投资产品或投资决策之依据,亦不应被诠释为专业意见。阅览本文件的人士或在作出任何投资决策前,应完全瞭解其风险以及有关法律、赋税及会计的特点及後果,并根据个人的情况决定投资是否切合个人的投资目标,以及能否承担有关风险,必要时应寻求适当的专业意见。投资涉及风险。敬请投资者注意,证券及投资的价值可升亦可跌,过往的表现不一定可以预示日後的表现。在若干国家,传阅及分派本档的方式可能受法律或规例所限制。获取本档的人士须知悉及遵守该等限制。
大部分微信公众号研报本站已有pdf详细完整版:https://www.wkzk.com/report/(可搜索研报标题关键词或机构名称查询原报告)
郑重声明:悟空智库网发布此信息的目的在于传播更多信息,与本站立场无关,不构成任何投资建议。