Shift Technologies, Inc.(SFT)_个股概要_股票价格_最新资讯_行情走势_历史数据 - 老虎社区

应交易所要求，美/港股行情延时15分钟展示，下载APP即可免费查看实时行情。

关闭

SFT Shift Technologies, Inc.

退市 10-18 16:00:00 EDT

0.1703

+0.0000

0.00%

最高0.1703	最低0.1703	成交量0.00
今开0.1703	昨收0.1703	日振幅0.00%
总市值289.49万	流通市值214.35万	总股本1,700万
成交额3,939万	换手率0.00%	流通股本1,259万
市净率--	ROE--	每股收益0.00
52周最高--	52周最低--	市盈率--
股息--	股息收益率--	ROA--

数据加载中...

北京优虎网络科技有限公司

{"stockData":{"symbol":"SFT","market":"US","secType":"STK","nameCN":"Shift Technologies, Inc.","latestPrice":0.1703,"timestamp":1697659200000,"preClose":0.1703,"halted":4,"volume":0,"delay":0,"floatShares":12586491,"shares":16999021,"eps":-10.42588,"marketStatus":"退市","change":0,"latestTime":"10-18 16:00:00 EDT","open":0.1703,"high":0.1703,"low":0.1703,"amount":39392485.363854,"amplitude":0,"askPrice":0,"askSize":0,"bidPrice":0,"bidSize":0,"shortable":3,"etf":0,"ttmEps":-10.42588,"tradingStatus":0,"nextMarketStatus":{"tag":"盘前交易","tradingStatus":1,"beginTime":1765789200000},"marketStatusCode":7,"adr":0,"exchange":"NASDAQ","adjPreClose":0.1703,"volumeRatio":0},"requestUrl":"/m/hq/s/SFT","defaultTab":"news","newsList":[{"id":"2586781841","title":"国产最强多模态宝座又易主？671B参数练就“火眼金睛”，基于DeepSeek打造","url":"https://stock-news.laohu8.com/highlight/detail?id=2586781841","media":"智东西","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2586781841?lang=zh_cn&edition=full","pubTime":"2025-11-28 18:41","pubTimestamp":1764326495,"startTime":"0","endTime":"0","summary":"基于Keye-VL-preview的预实验显示，Keye-Verifier提供的奖励信号，相对于基于规则匹配的奖励信号，","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://tech.ifeng.com/c/8oecXFyVuVl","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"ifeng_tech","symbols":["BK1575","LU0244354667.USD","LU1794554557.SGD","LU0348766576.USD","LU0348767384.USD","LU2097828805.USD","LU0593848301.USD","LU0572944931.SGD","LU0327786744.USD","LU0326950275.SGD","LU2097828714.EUR","81024","LU0051755006.USD","LU1720050803.USD","LU2097828631.EUR","LU1303224171.USD","LU0359201885.HKD","SFT","01024","LU1719994722.HKD","LU2097828557.USD","BK1095","BK1610","BK1608","KSTmain","LU0210526637.USD","LU0463099449.HKD","LU2097828474.EUR","LU0211977185.USD","LU1023057109.AUD","BK1590","BK4214","LU1251922891.USD","LU0307460666.USD","BK1615","LU1770034418.SGD","LU0359202008.SGD","LU0359201612.USD","LU0140636845.USD","BK1591","LU1188198961.HKD","LU0456827905.SGD","LU0588546209.SGD"],"gpt_icon":1},{"id":"2577843120","title":"刚刚，DeepSeek重要突破！大模型上下文紧箍咒打破","url":"https://stock-news.laohu8.com/highlight/detail?id=2577843120","media":"智东西","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2577843120?lang=zh_cn&edition=full","pubTime":"2025-10-21 11:06","pubTimestamp":1761015970,"startTime":"0","endTime":"0","summary":"目前，这一模型已在Hugging Face上开源，而介绍DeepSeek-OCR模型技术细节与背后理论的技术报告也已同步公开。DeepSeek-OCR团队称，他们此番开源的模型是对一种潜在解决方案的初步探索，即利用视觉模态作为文本信息的高效压缩媒介。DeepSeek-OCR的架构分为两部分。当DeepEncoder训练完成后，DeepSeek-OCR团队使用多模态数据和纯文本数据，采用流水线并行策略来训练完整的模型。实验结果显示，在10×压缩率下，DeepSeek-OCR的解码精度可达约97%。","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://tech.ifeng.com/c/8nd8NmIG6UZ","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"fenghuang_stock","symbols":["SFT","BK4214"],"gpt_icon":0},{"id":"2560295411","title":"智谱推出 AutoGLM 2.0：手机 Agent 的“Manus 时刻”？","url":"https://stock-news.laohu8.com/highlight/detail?id=2560295411","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2560295411?lang=zh_cn&edition=full","pubTime":"2025-08-20 12:34","pubTimestamp":1755664440,"startTime":"0","endTime":"0","summary":"作者｜连冉编辑｜郑玄8 月 18 日，智谱正式发布了新的 ToC 产品 AutoGLM 2.0——一个手机通用 Agent。现在，当 AutoGLM 再次进入公众视野，情形已有所不同。AutoGLM 的新方案是用“云端原生”取代“本地镜像”。在与团队的交流中，极客公园了解到 AutoGLM 的产品哲学可以被提炼为“3A 原则”。这三大原则，共同构成了 AutoGLM 对一个成熟 Agent 形态的定义，也解释了其当前产品架构。据智谱披露，通过在线强化学习，AutoGLM 的任务成功率相较于冷启动阶段提升了 165%，超过 66% 的","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-08-20/doc-infmrakf1841118.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4214","SFT"],"gpt_icon":0},{"id":"2558652406","title":"智谱GLM-4.5完整技术报告：提出三个通用模型关键能力，公开12项测试成绩","url":"https://stock-news.laohu8.com/highlight/detail?id=2558652406","media":"智东西","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2558652406?lang=zh_cn&edition=full","pubTime":"2025-08-11 18:07","pubTimestamp":1754906826,"startTime":"0","endTime":"0","summary":"编译 | 陈骏达编辑 | 云鹏智东西8月11日报道，近日，智谱发布了其最新一代旗舰模型GLM-4.5的完整技术报告。智东西此前已对GLM-4.5的能力进行了介绍与测试，在技术报告中，智谱进一步分享了这款模型在预训练、中期训练和后训练阶段进行的创新。值得一提的是，智谱还计划在今晚开源GLM-4.5系列的新模型，名为GLM-4.5V，或为一款视觉模型。SFT之后，GLM-4.5又进行了强化学习训练。","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://tech.ifeng.com/c/8ljg4SI98PH","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"fenghuang_stock","symbols":["BK4017","SFT","LU0498741890.SGD","LU0498741114.HKD","AGI","LU0368265764.SGD","LU0496367417.USD","LU0055631609.USD","BK4214"],"gpt_icon":0},{"id":"2558687160","title":"陈天桥联手清华教授代季峰首发最强开源AI模型项目，全力打造下一个DeepSeek","url":"https://stock-news.laohu8.com/highlight/detail?id=2558687160","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2558687160?lang=zh_cn&edition=full","pubTime":"2025-08-11 07:39","pubTimestamp":1754869140,"startTime":"0","endTime":"0","summary":"有报道称，陈天桥对代季峰领衔的这家新 AI 创业公司寄予厚望，还承诺，盛大内部孵化的所有AI企业的一半利润将分给团队。如今，AI大牛代季峰再度“出山”，与创新企业家、慈善家、天桥脑科学研究院创始人陈天桥联手筹备一家新的AI创业公司，目标是打造下一个OpenAI，第二个DeepSeek，将围绕AGI展开基础性研究，首个项目就是MiroMind Open Deep Research。目前，代季峰的MiroMind团队已经对外开放MiroMind ODR项目Demo进行体验。","market":"sh","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-08-11/doc-infkqfti6088263.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"1","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4017","BK4214","AGI","LU0498741114.HKD","LU0498741890.SGD","SFT","LU0496367417.USD","LU0055631609.USD","LU0368265764.SGD"],"gpt_icon":0},{"id":"2557153503","title":"站在DeepSeek肩膀上，小红书开源首款多模态模型：看懂表情包与数学题，一手实测","url":"https://stock-news.laohu8.com/highlight/detail?id=2557153503","media":"智东西","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2557153503?lang=zh_cn&edition=full","pubTime":"2025-08-07 12:30","pubTimestamp":1754541040,"startTime":"0","endTime":"0","summary":"今年6月6日，小红书开源了其首款大语言模型，并在之后开源了用于OCR的专用模型，以及视觉、奖励模型等前沿方向的研究成果。VLM预训练在这一阶段，hi lab将视觉编码器与DeepSeek V3联合训练，使用大规模、多样化的多模态数据集，主要包括跨模态互译数据和跨模态融合数据。","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://tech.ifeng.com/c/8lcepKwpv4H","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"fenghuang_stock","symbols":["BK4214","SFT"],"gpt_icon":0},{"id":"2553261454","title":"复旦联合南洋理工提出基于视觉Grounding的多轮强化学习框架MGPO","url":"https://stock-news.laohu8.com/highlight/detail?id=2553261454","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2553261454?lang=zh_cn&edition=full","pubTime":"2025-07-21 13:46","pubTimestamp":1753076760,"startTime":"0","endTime":"0","summary":"为解决这一问题，复旦大学、南洋理工大学的研究者提出一种基于视觉 Grounding 的多轮强化学习方法 MGPO，使 LMM 能在多轮交互中根据问题，自动预测关键区域坐标，裁剪子图像并整合历史上下文，最终实现高分辨率图像的精准推理。但传统视觉 Grounding 模型需依赖大量 Grounding 标注进行训练，而此类标注成本较高。","market":"fut","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-21/doc-infhfezw4892690.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["LU0006061336.USD","BK4214","BK4202","SFT","RL","BK4585","BK4588"],"gpt_icon":0},{"id":"2552443983","title":"AI打假AI，拿下SOTA丨厦大&腾讯优图","url":"https://stock-news.laohu8.com/highlight/detail?id=2552443983","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2552443983?lang=zh_cn&edition=full","pubTime":"2025-07-20 15:00","pubTimestamp":1752994800,"startTime":"0","endTime":"0","summary":"具体方法是AIGI-HolmesAI生成图像检测方法，由厦门大学多媒体可信感知与高效计算教育部重点实验室和腾讯优图团队带来。团队针对上述问题，通过AIGI-Holmes给出解决方案。其中CLIP用于检测high-level semantic缺陷，而NPR则用于检测low-level artfacts，分别在Holmes-set上进行LoRA微调和全参微调。具体来说，团队在三个AIGI检测的数据集上评估了检测能力，包括AIGCDetect-Benchmark、AntiFakePrompt，并且额外采集了10种SOTA生成模型的图片构建了第三个benchmark，用于测试模型在未见过的生成方法上的泛化能力。","market":"sg","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-20/doc-infhcaau4129839.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4214","SFT"],"gpt_icon":0},{"id":"2551975105","title":"OpenAI联合创始人揭秘AI进化新方向：让模型学会和人类一样反思","url":"https://stock-news.laohu8.com/highlight/detail?id=2551975105","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2551975105?lang=zh_cn&edition=full","pubTime":"2025-07-14 11:25","pubTimestamp":1752463500,"startTime":"0","endTime":"0","summary":"Karpathy 觉得，RL 缺少这种类似人类反思的机制，而这可能是 LLMs 未来进化的关键。Karpathy 用“second nature”来形容人类通过反思逐渐掌握技能的过程。Karpathy 认为，AI 应该也有类似机制，尤其是像 LLMs 这样有强大语言能力和上下文学习能力的模型。Karpathy 认为，RL 确实比监督微调更“苦涩”，而且还会带来更多性能提升。Karpathy 的设想是：如果能让模型自己总结经验教训，并在实践中不断优化，可能会开启 AI 智能的新篇章。","market":"other","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-14/doc-inffmawy3285895.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["RL","BK4202","BK4585","SFT","BK4214","LU0006061336.USD","BK4588"],"gpt_icon":0},{"id":"2550614919","title":"豆蔻妇科大模型再突破：钉钉行业训练平台+精标数据SFT ，准确率从 77.1%上升至 90.2%","url":"https://stock-news.laohu8.com/highlight/detail?id=2550614919","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2550614919?lang=zh_cn&edition=full","pubTime":"2025-07-10 15:39","pubTimestamp":1752133140,"startTime":"0","endTime":"0","summary":"豆蔻妇科大模型的模型调优经历了两个关键优化阶段：以下是豆蔻妇科大模型从第一个版本的准确率77.1%，通过进一步的SFT后，准确率达到90.2%我们团队的一些方法和心得，供大家参考，欢迎留言讨论。例如，针对 “患者出现阴道出血症状，诊断为宫颈病变” 等诊断结论，依据医学知识库中的关联规则，自动评估其逻辑合理性与临床可行性。在这一阶段的调优过程中，钉钉企业专属AI平","market":"other","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-10/doc-infeypam3762985.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"1","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["LU0048580855.USD","HBBD.SI","LU0516423091.SGD","LU0737861772.HKD","LU0488056044.USD","LU0819123356.HKD","LU2242644610.SGD","BK1584","LU0084288322.USD","LU1323998911.USD","LU0293314216.USD","LU0588546209.SGD","LU0831103253.SGD","IE0034224299.USD","LU0163747925.USD","LU0516423174.USD","LU0611395673.USD","LU0449515922.USD","LU0345776255.USD","LU1201861165.SGD","BK1517","LU0029874905.USD","LU0640798160.USD","LU1051768304.USD","LU0348735423.USD","LU0029875118.USD","LU1808992512.USD","LU0823397285.USD","SG9999001689.USD","ALBmain","LU0329678337.USD","SG9999014674.SGD","LU0315178854.USD","89988","LU0979878070.USD","BK1615","LU0499858602.USD","SFT","LU0348805143.USD","LU0449509016.USD","LU0577902371.SGD","LU1568876335.HKD","LU0348784397.USD","LU0140636845.USD","BK1521","LU0593848301.USD","LU0642271901.SGD","LU1515016050.SGD"],"gpt_icon":1},{"id":"2548865185","title":"智谱再融10亿！获上海国资押注，开源视觉大模型，能解说球赛，还会玩手机","url":"https://stock-news.laohu8.com/highlight/detail?id=2548865185","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2548865185?lang=zh_cn&edition=full","pubTime":"2025-07-02 15:40","pubTimestamp":1751442000,"startTime":"0","endTime":"0","summary":"开源之外，智谱还在今天举行的智谱开放平台产业生态大会上宣布，该公司获得浦东创投集团和张江集团联合战略投资，总额10亿元。目前，开源社区缺乏一种在广泛任务范围内持续超越传统同类参数规模非推理模型的多模态推理模型。在视觉编码器部分，智谱将原始的二维卷积替换为三维卷积，尤其适用于视频理解，有效提升了处理效率。","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-02/doc-infeanhk2500975.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"1","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4202","BK4588","RL","SFT","LU0006061336.USD","BK4585","BK4214"],"gpt_icon":0},{"id":"2548597918","title":"10B级模型SOTA，超8倍参数“大”模型，智谱开源GLM-4.1V-Thinking","url":"https://stock-news.laohu8.com/highlight/detail?id=2548597918","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2548597918?lang=zh_cn&edition=full","pubTime":"2025-07-02 12:22","pubTimestamp":1751430120,"startTime":"0","endTime":"0","summary":"智谱正式发布 GLM-4.1V-Thinking 系列模型，并率先开源GLM-4.1V-9B-Thinking，标志着智谱 GLM 视觉大模型向高阶认知迈出了关键一步。在 18 项权威评测中，GLM-4.1V-9B-Thinking 的表现已可比肩甚至超越参数量高达 72B 的 Qwen2.5-VL-72B，充分展示出结构设计与训练策略的先进性与效率。模型原理1. 模型架构GLM-4.1V-Thinking 模型架构由三个核心模块组成：视觉编码器、多层感知机适配器以及语言解码器。2训练流程GLM-4.1V-Thinking 的训练过程分为三个阶段：预训练、监督微调 和 强化学习。","market":"fut","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-07-02/doc-infczzsq2682912.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4214","BK4588","LU0006061336.USD","SFT","BK4202","RL","BK4585"],"gpt_icon":0},{"id":"2547018030","title":"AI字幕慢半拍，不知道大家笑什么？新方法让同传性能直逼离线翻译","url":"https://stock-news.laohu8.com/highlight/detail?id=2547018030","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2547018030?lang=zh_cn&edition=full","pubTime":"2025-06-30 21:29","pubTimestamp":1751290140,"startTime":"0","endTime":"0","summary":"来自香港中文大学、字节跳动 Seed 和斯坦福大学的研究团队联手提出了一种面向同声传译的序贯策略优化框架 该方法将同传任务巧妙地建模为序贯决策过程，通过优化完整的决策序列，显著提升了翻译质量，同时有效控制了延迟，其性能直逼、甚至在某些方面超越了同等大小的离线翻译模型。本文将 SeqPO-SiMT 的实时同传结果与多个高性能模型的离线翻译结果进行对比。","market":"us","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-30/doc-infcwksa9651178.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4214","SFT"],"gpt_icon":0},{"id":"2545222402","title":"强化学习新发现：无需数学样本，仅游戏训练AI推理大增","url":"https://stock-news.laohu8.com/highlight/detail?id=2545222402","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2545222402?lang=zh_cn&edition=full","pubTime":"2025-06-24 16:46","pubTimestamp":1750754760,"startTime":"0","endTime":"0","summary":"最近，强化学习领域出现了一个颠覆性发现：研究人员不再需要大量数学训练样本，仅仅让 AI 玩简单游戏，就能显著提升其数学推理能力。此前已有研究发现，即使不提供标准答案，仅用数学问题进行强化学习也能提高模型性能，这让人们开始重新思考强化学习的训练方式。执行后会从环境获得奖励，用于进行强化学习。两款游戏的设计哲学互补：贪吃蛇主要提升 2D 坐标相关的数学表现，旋转游戏则更适合角度和长度推理。","market":"hk","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-24/doc-infcemnh8974111.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4202","BK4585","BK4588","LU0006061336.USD","BK4214","SFT","RL"],"gpt_icon":0},{"id":"2545049026","title":"我在哪？要去哪？字节跳动Astra双模型架构助力机器人自由导航","url":"https://stock-news.laohu8.com/highlight/detail?id=2545049026","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2545049026?lang=zh_cn&edition=full","pubTime":"2025-06-23 17:44","pubTimestamp":1750671840,"startTime":"0","endTime":"0","summary":"为了突破传统导航系统的瓶颈，字节跳动研发了一种创新的双模型架构 Astra。图1: Astra模型概述二、Astra 双模型架构揭秘，赋能机器人高效导航1. Astra-Global：全局定位的智慧大脑Astra-Global 作为 Astra 架构中的重要组成部分，犹如智慧大脑，承担着关键的低频任务，即自我定位和目标定位。在实际定位过程中，Astra-Global 的自定位与目标定位功能展现出独特的优势。图3: Astra-Local 架构4D 时空编码器是 Astra-Local 的基础组件，它旨在取代传统移动性堆栈中的感知和预测模块。","market":"other","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-23/doc-infcahph9900776.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["SFT","BK4214"],"gpt_icon":0},{"id":"2544993277","title":"沉迷贪吃蛇，7B小模型竟变身“数学天才”！几何推理碾压GPT-4o","url":"https://stock-news.laohu8.com/highlight/detail?id=2544993277","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2544993277?lang=zh_cn&edition=full","pubTime":"2025-06-17 13:43","pubTimestamp":1750138980,"startTime":"0","endTime":"0","summary":"通过让7B参数的多模态模型玩贪吃蛇和3D旋转等街机游戏，AI不仅掌握了游戏技巧，还培养出强大的跨领域推理能力，在数学、几何等复杂任务上击败GPT-4o等顶级模型。但让人没想到的是，通过游戏的训练，这个AI还可以成为一位“数学天才”！结果显示，一个沉迷于街机游戏的7B参数MLLM，竟然在复杂的数学和几何推理任务上，一举击败了GPT-4o这样的顶级闭源大模型。","market":"sh","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-17/doc-infakiik8699086.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["SFT","BK4214"],"gpt_icon":0},{"id":"2544934463","title":"Kimi新模型拿下代码开源SOTA，仅仅72B，发布即开源","url":"https://stock-news.laohu8.com/highlight/detail?id=2544934463","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2544934463?lang=zh_cn&edition=full","pubTime":"2025-06-17 12:58","pubTimestamp":1750136280,"startTime":"0","endTime":"0","summary":"深夜，沉寂已久的Kimi突然发布了新模型——开源代码模型Kimi-Dev，在SWE-bench Verified上以60.4%的成绩取得开源SOTA。Kimi-Dev采用MIT协议，目前权重和代码均已发布，还有网友制作的量化版本也已在抱抱脸上线。具体来说，Kimi团队收集了数百万个GitHub issue和PR提交，目的是让Kimi-Dev-72B能够学习人类开发人员如何推理并解决GitHub问题。","market":"other","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-17/doc-infakazm6738869.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["BK4214","SFT"],"gpt_icon":0},{"id":"2543678781","title":"SFT+RL双管齐下：ReasonGen-R1如何破解文生图“指令不遵”难题？","url":"https://stock-news.laohu8.com/highlight/detail?id=2543678781","media":"市场资讯","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2543678781?lang=zh_cn&edition=full","pubTime":"2025-06-16 17:22","pubTimestamp":1750065720,"startTime":"0","endTime":"0","summary":"ReasonGen-R1 使得自回归图像生成模型可以端到端地在输出图片之前先进行文本“思考”，大幅提升了基座模型的语义遵循能力，并在多个语义指标上取得突破。目前，ReasonGen-R1 已全面开源。方法概览ReasonGen-R1 的训练包括两个核心阶段：监督微调阶段以及强化学习阶段。","market":"fut","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/stock/t/2025-06-16/doc-infahksh9599543.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["LU0006061336.USD","BK4214","BK4202","SFT","RL","BK4585","BK4588"],"gpt_icon":0},{"id":"2512998724","title":"干货满满！关于Deepseek，基金经理最关心的事······","url":"https://stock-news.laohu8.com/highlight/detail?id=2512998724","media":"新浪基金","labels":[],"top":-1,"itemType":null,"share":"https://www.laohu8.com/m/news/2512998724?lang=zh_cn&edition=full","pubTime":"2025-02-20 14:28","pubTimestamp":1740032880,"startTime":"0","endTime":"0","summary":"在最新一期泰客Talk《Deepseek的真相和谎言》中，基金经理于腾达和券商分析师童飞分享了关于Deepseek对二级市场投资的影响的观点。干货满满，整理如下——。　　Deepseek做的是一件科学的工作，但是投资不只是一件科学的数据性的工作。","market":"other","thumbnail":null,"type":0,"news_type":0,"thumbnails":[],"rights":null,"property":[],"language":"zh","translate_title":"","themeId":null,"theme_name":"","theme_type":"","isJumpTheme":false,"source_url":"https://finance.sina.com.cn/money/fund/jjh/2025-02-20/doc-inemchxk2430529.shtml","is_publish_highlight":false,"source_rank":0,"column":"","sentiment":"0","news_top_title":null,"news_tag":"","news_rank":0,"length":0,"strategy_id":0,"source":"sina","symbols":["LU1917777945.USD","NVDA","LU1119994496.HKD","IE00BJJMRX11.SGD","HK0000320264.USD","LU1366192091.USD","SNVD.UK","NVDY","LU0417517546.SGD","LU0345769631.USD","2NVD.UK","RL","BK4532","NVDU","IE00BMPRXN33.USD","SG9999004303.SGD","LU2097344357.USD","LU1718418525.SGD","LU0943347566.SGD","LU0056508442.USD","LU2065171402.SGD","BK4605","LU0082616367.USD","LU1923623000.USD","NVD","LU0316494557.USD","SGXZ51526630.SGD","LU2471134796.USD","SG9999015986.USD","SFT","NVD2.UK","IE00BN29S564.USD","NVD3.UK","LU1720051017.SGD","NVDS","LU0061475181.USD","IE00BYXW3230.USD","NVDD","NVIW.SI","NVDX","LU1261432733.SGD","LU0787776722.HKD","BK4549","NVDS.UK","LU0048584097.USD","LU0265550359.USD","LU0094547139.USD","LU1712237335.SGD","3NVD.UK"],"gpt_icon":1}],"profile":{"websiteUrl":"http://www.shift.com","stockEarnings":[{"period":"1week","weight":0},{"period":"1month","weight":0},{"period":"3month","weight":0},{"period":"6month","weight":0},{"period":"1year","weight":0},{"period":"ytd","weight":0}],"compareEarnings":[{"period":"1week","weight":-0.0057},{"period":"1month","weight":-0.0024},{"period":"3month","weight":0.037},{"period":"6month","weight":0.142},{"period":"1year","weight":0.1281},{"period":"ytd","weight":0.1633}],"compareStock":{"symbol":"SPY","name":"标普500ETF"},"description":"Shift Technologies, Inc.于2018年3月13日在特拉华州成立。该公司是一家以消费者为中心的全渠道零售商，通过利用其端到端电子商务平台和零售地点提供技术驱动、无忧的客户体验，改变了二手车行业。该公司在加州和俄勒冈州拥有三个市场，客户可以在这里试驾车辆，并在销售过程中的任何一个环节与销售伙伴合作。","yearOnYearQuotes":[{"month":1,"riseRate":0.5,"avgChangeRate":0.008105},{"month":2,"riseRate":0.75,"avgChangeRate":0.000831},{"month":3,"riseRate":0.25,"avgChangeRate":-0.111504},{"month":4,"riseRate":0.75,"avgChangeRate":-0.022745},{"month":5,"riseRate":0.25,"avgChangeRate":-0.075173},{"month":6,"riseRate":0.75,"avgChangeRate":0.134273},{"month":7,"riseRate":0.2,"avgChangeRate":0.046683},{"month":8,"riseRate":0.4,"avgChangeRate":-0.083325},{"month":9,"riseRate":0.4,"avgChangeRate":-0.005575},{"month":10,"riseRate":0,"avgChangeRate":-0.28076},{"month":11,"riseRate":0.25,"avgChangeRate":-0.186599},{"month":12,"riseRate":0.25,"avgChangeRate":-0.20848}],"exchange":"NASDAQ","name":"Shift Technologies, Inc.","nameEN":"Shift Technologies, Inc."},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":false,"deviceId":"web-server-community-laohu8-v3","version":"4.35.11","shortVersion":"4.35.11","platform":"web","vendor":"web","appName":"laohu8","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"pagemeta":{"title":"Shift Technologies, Inc.(SFT)_个股概要_股票价格_最新资讯_行情走势_历史数据","description":"美港股上老虎。老虎社区提供Shift Technologies, Inc.(SFT)今日价格,行情走势,历史数据,股票概要及实时的新闻资讯,近期大事等重要参考决策数据。","keywords":"Shift Technologies, Inc.,SFT,Shift Technologies, Inc.股票,Shift Technologies, Inc.股票老虎,Shift Technologies, Inc.股票老虎国际,Shift Technologies, Inc.行情,Shift Technologies, Inc.股票行情,Shift Technologies, Inc.股价,Shift Technologies, Inc.股市,Shift Technologies, Inc.股票价格,Shift Technologies, Inc.股票交易,Shift Technologies, Inc.股票购买,Shift Technologies, Inc.股票实时行情,购买美股,购买港股,港股开户,美股开户,美股交易,港股交易,开通美港股账户,老虎国际行情","social":{"og_title":"Shift Technologies, Inc.(SFT)_个股概要_股票价格_最新资讯_行情走势_历史数据","og_description":"美港股上老虎。老虎社区提供Shift Technologies, Inc.(SFT)今日价格,行情走势,历史数据,股票概要及实时的新闻资讯,近期大事等重要参考决策数据。","og_image":"https://static.tigerbbs.com/a0b84b8debbbce3b4440f7fdd5eed2e7"}}}