China’s AI labs must accelerate development, says Huawei chair - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
华为

China’s AI labs must accelerate development, says Huawei chair

Chinese tech executive’s comments contrast with Silicon Valley calls for slowdown amid rising concerns about technology’s existential risks
00:00

{"text":[[{"start":7.76,"text":"Huawei’s chair has urged Chinese AI companies to accelerate development to catch up with frontier US labs, contrasting with calls in Silicon Valley for a slowdown amid rising concerns about the existential risks posed by the technology."}],[{"start":22.28,"text":"Chinese researchers needed to “increase the speed of development so they can also see the dangers of AI development” like their US counterparts, said Eric Xu, the company’s rotating chair, at a press conference on Thursday, adding that once those risks emerged, the industry would need to “strike a balance between driving development and risks”."}],[{"start":41.84,"text":"The Chinese tech executive’s comments came just days after Anthropic and OpenAI called for a co-ordinated slowdown in the technology’s development following days of stark warnings over its safety."}],[{"start":53.48,"text":"They also came as Huawei unveiled updates to its AI hardware plans. The Chinese tech giant is seeking to build an alternative to Nvidia’s dominant AI computing platform, as Washington continues to restrict exports of the most advanced AI chips to China."}],[{"start":69.08,"text":"Xu acknowledged Huawei remained “not as advanced” as leading US rivals but said the unpredictable export controls had pushed Chinese customers towards domestic technology."}],[{"start":79.16,"text":"“We can’t accept a destiny that we cannot control,” he said. “No matter if it’s the Chinese government or Huawei, we are pushing for full self-sufficiency for chips and the entire supply chain around semiconductors.”"}],[{"start":92.28,"text":"Huawei has become increasingly confident in publicly outlining its AI ambitions, taking the unusual step last September of unveiling a detailed multiyear roadmap for its Ascend chips."}],[{"start":103.96,"text":"On Thursday, the company said its Ascend 960 DT chip, which targets training of AI models, would launch in the first quarter of 2027, while its chip for inference — the process of AI models running queries — would come in the third quarter. It added they would deliver twice the computing performance of its current flagship product."}],[{"start":123.16,"text":"At the heart of Huawei’s effort to challenge Nvidia is its networking technology, which it says allows it to build larger clusters than its US rival without compromising bandwidth and stability."}],[{"start":133.66,"text":"The approach is central to Huawei’s effort to compensate for the weaker performance of its individual AI processors, as US export controls have prevented it from using leading-edge contractors such as Taiwan’s TSMC."}],[{"start":145.2,"text":"This has forced the company to rely more heavily on improvements in networking and system architecture to combine large numbers of Ascend processors into powerful AI computing systems."}],[{"start":157.56,"text":"Huawei’s data centre product built around the chips will connect as many as 4,096 Ascend processors, compared with 1,024 in the previous generation, allowing customers to train and run larger AI models. The company has struggled to get AI labs to successfully use its chips for training large models."}],[{"start":178.02,"text":"Xu estimated that Ascend had overtaken Nvidia in market share in China while acknowledging that reliable estimates were difficult to obtain."}],[{"start":186.32,"text":"Heng Liao, chief scientist for Huawei’s semiconductor division, said changes in how increasingly large and complex AI models are trained were also eroding the advantage of Nvidia’s CUDA software platform, which has long helped lock developers into the US chipmaker’s graphics processing units."}],[{"start":203.04,"text":"“CUDA is not as important as it was two years ago,” said Liao, pointing to the emergence of new programming tools and approaches for training AI models. “The software gap is greatly diminished.”"}],[{"start":214.84,"text":"Xu said at least six or seven Chinese AI labs were aiming over the next two years to train models containing 10tn to 40tn parameters, a significant increase from current model sizes in China."}],[{"start":231.78,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1789687477_2552.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

OpenAI上市推迟,令软银数据中心IPO的风险增高

OpenAI将持有SB Energy上市后的大量股份,后者估值500亿美元的IPO将检验投资者对孙正义AI赌注的兴趣。

厄尔尼诺现象会成为欧洲能源市场的救星吗?

欧洲已接连遭受能源危机的重创,而目前的策略却是寄希望于天气不要太冷。

Lex专栏:欧洲可通过合并交易所吸引更多股票交易

一个规模更大的欧洲交易所,可以更好地吸引寻求在高度集中于科技行业的美股之外实现多元化配置的美国投资者。

马克•卡尼如何打好了一手烂牌?

自特朗普重返白宫以来,很少有哪个国家像加拿大一样频频成为他的攻击目标。不过到目前为止,加拿大总理仍然在相当娴熟地应对着各种局面。

新的工作岗位从哪里来?

凯恩斯:延续了数百年的专业分工趋势,或许正面临逆转。

AI时代的苹果信任溢价

阿姆斯特朗:在一个充斥AI风险的市场中,苹果获得溢价的原因并不只是没有AI风险;苹果出售的是这个世界越来越缺乏的信任。
设置字号×
最小
较小
默认
较大
最大
分享×