7*24

Hot topic

Market

Activities

According to monitoring by Dongcha Beating, Zhou Chang, head of ByteDance's multimodal division, has recently expanded his management scope again, with the Seed Robotics team, previously led by Li Hang, now reporting to him for over a month. Meanwhile, ByteDance is recruiting a head of embodied intelligence technology responsible for the overall planning of the robotics business, targeting technical leaders from top embodied intelligence startups, with a job level benchmarked against Alibaba's P10 to P11 L8 levels. The integration of the Robotics team at ByteDance signifies a shift towards centralized management of R&D resources in the robotics direction. The combination of multimodal interaction, world models, and embodied intelligence business is conducive to the deep reuse of technical needs and data resources in the robotics sector. In the previous structure, the hardware, data, and model development for robotics were handled by different teams, leading to overlaps in algorithm development. After the adjustment, robotics will serve as a vehicle for large models to be implemented in the physical world, providing real interaction data from scenarios such as warehouses, factories, and homes for multimodal models. As the core R&D leader of Seed, Zhou Chang has seen his responsibilities expand multiple times since joining ByteDance from Alibaba's Damo Academy in 2024. Beyond multimodal interaction and world models, he has also taken over visual generation businesses such as the text-to-image model Seedream and the text-to-video model Seedance, following the departure of visual foundational model leader Feng Jia and the leave of visual multimodal generation leader Yang Jianchao. The Robotics team, now under Zhou Chang's management, was originally part of ByteDance's AI Lab and was integrated into the Seed system in 2025. The former head, Li Hang, now reports to the head of academic cooperation, Jin Xiaowei, and serves as a consultant for academic collaboration. In addition to ByteDance concentrating resources to build a technological closed loop, OpenAI CEO Sam Altman also announced on June 1 that the OpenAI robotics team is recruiting full-stack hardware, operations, systems, and machine learning engineers, planning to develop socially useful robots through programming and manufacturing, gradually moving from short-term infrastructure building to a long-term vision of personal assistants.

[ByteDance Restructures Seed Robotics Team: Zhou Chang Takes Over Embodied Intelligence, L8 Executive Recruitment Targets Startups]According to monitoring by Dongcha Beating, Zhou Chang, head of ByteDance's multimodal division, has recently expanded his management scope again, with the Seed Robotics team, previously led by Li Hang, now reporting to him for over a month. Meanwhile, ByteDance is recruiting a head of embodied intelligence technology responsible for the overall planning of the robotics business, targeting technical leaders from top embodied intelligence startups, with a job level benchmarked against Alibaba's P10 to P11 L8 levels. The integration of the Robotics team at ByteDance signifies a shift towards centralized management of R&D resources in the robotics direction. The combination of multimodal interaction, world models, and embodied intelligence business is conducive to the deep reuse of technical needs and data resources in the robotics sector. In the previous structure, the hardware, data, and model development for robotics were handled by different teams, leading to overlaps in algorithm development. After the adjustment, robotics will serve as a vehicle for large models to be implemented in the physical world, providing real interaction data from scenarios such as warehouses, factories, and homes for multimodal models. As the core R&D leader of Seed, Zhou Chang has seen his responsibilities expand multiple times since joining ByteDance from Alibaba's Damo Academy in 2024. Beyond multimodal interaction and world models, he has also taken over visual generation businesses such as the text-to-image model Seedream and the text-to-video model Seedance, following the departure of visual foundational model leader Feng Jia and the leave of visual multimodal generation leader Yang Jianchao. The Robotics team, now under Zhou Chang's management, was originally part of ByteDance's AI Lab and was integrated into the Seed system in 2025. The former head, Li Hang, now reports to the head of academic cooperation, Jin Xiaowei, and serves as a consultant for academic collaboration. In addition to ByteDance concentrating resources to build a technological closed loop, OpenAI CEO Sam Altman also announced on June 1 that the OpenAI robotics team is recruiting full-stack hardware, operations, systems, and machine learning engineers, planning to develop socially useful robots through programming and manufacturing, gradually moving from short-term infrastructure building to a long-term vision of personal assistants.

ByteDance Restructures Seed Robotics Team: Zhou Chang Takes Over Embodied Intelligence, L8 Executive Recruitment Targets Startups

According to monitoring by Dongcha Beating, large model agents often evolve by updating external "equipment" (Harness, which includes prompts, memory, skills, and tools). A recent paper titled "Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents" published by institutions including Penn State University, UCSC, and Amazon, for the first time decouples the evolution process into two dimensions: the "equipment update" of the evolver and the "equipment benefit" at the execution end. Cross-testing indicates that the ability to update equipment shows a significant "flattening" characteristic in the model's foundational capabilities. The maximum benefit difference from updating equipment across different models is only 3.1%, and even the updated skills produced by the 9B-sized Qwen3.5-9B are structurally highly equivalent to the flagship Claude Opus 4.6. This suggests that high-cost models are not necessary for developing self-evolving systems. Conversely, the ability of agents to benefit from equipment shows a "non-monotonic" trend. Top models are nearing their performance ceiling, while weaker models (like Qwen3-32B), despite having the most room for improvement, actually gain the least. The research identifies two major failure modes in weaker models. First, there is "equipment activation failure," where the skill loading rate of weaker models in the SkillsBench benchmark is only 25.1%, while strong models have a loading rate of about 96%. Second, there is "equipment adherence failure," where the instruction adherence of weaker models drops sharply from 0.52 at the beginning of loading to 0.13 as long-range execution trajectories unfold. These cutting-edge findings resonated strongly with renowned AI researcher Elvis Sar (@omarsar0), who noted that he observed a completely consistent phenomenon in his experiments with coding agents and long-range tasks: more powerful models do not always evolve superior agents. This paper provides important guidance for the design of agent system architectures, indicating that computational power budgets should be skewed towards execution agents and that training should focus on enhancing autonomous equipment awakening and long-range instruction adherence capabilities.

[9B Small Model Self-Updating Agent Skills Match Claude's Flagship Model Performance]According to monitoring by Dongcha Beating, large model agents often evolve by updating external "equipment" (Harness, which includes prompts, memory, skills, and tools). A recent paper titled "Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents" published by institutions including Penn State University, UCSC, and Amazon, for the first time decouples the evolution process into two dimensions: the "equipment update" of the evolver and the "equipment benefit" at the execution end. Cross-testing indicates that the ability to update equipment shows a significant "flattening" characteristic in the model's foundational capabilities. The maximum benefit difference from updating equipment across different models is only 3.1%, and even the updated skills produced by the 9B-sized Qwen3.5-9B are structurally highly equivalent to the flagship Claude Opus 4.6. This suggests that high-cost models are not necessary for developing self-evolving systems. Conversely, the ability of agents to benefit from equipment shows a "non-monotonic" trend. Top models are nearing their performance ceiling, while weaker models (like Qwen3-32B), despite having the most room for improvement, actually gain the least. The research identifies two major failure modes in weaker models. First, there is "equipment activation failure," where the skill loading rate of weaker models in the SkillsBench benchmark is only 25.1%, while strong models have a loading rate of about 96%. Second, there is "equipment adherence failure," where the instruction adherence of weaker models drops sharply from 0.52 at the beginning of loading to 0.13 as long-range execution trajectories unfold. These cutting-edge findings resonated strongly with renowned AI researcher Elvis Sar (@omarsar0), who noted that he observed a completely consistent phenomenon in his experiments with coding agents and long-range tasks: more powerful models do not always evolve superior agents. This paper provides important guidance for the design of agent system architectures, indicating that computational power budgets should be skewed towards execution agents and that training should focus on enhancing autonomous equipment awakening and long-range instruction adherence capabilities.

9B Small Model Self-Updating Agent Skills Match Claude's Flagship Model Performance

On June 2, Tiger International announced that in order to implement industry regulatory requirements during a two-year concentrated rectification period and to promote the standardized development of cross-border securities business, it will make corresponding adjustments to the services for existing investor accounts in China. Starting from June 12, 2026, Beijing time, for trading services within the country, new and additional trades for all varieties, including stocks, will be suspended, with only sell and close operations supported. For fund transfer services within the country, fund inflows will be suspended, while outflows will remain normal to ensure the safety of customer funds. Tiger stated that this adjustment will not affect the services provided to existing investors overseas, nor will it impact the safety of existing assets for all customers. Customers can continue to query their accounts, hold, and sell existing positions normally. (Securities Times)

[Tiger International: Suspension of New and Additional Trades for Existing Investor Accounts in China Starting June 12]On June 2, Tiger International announced that in order to implement industry regulatory requirements during a two-year concentrated rectification period and to promote the standardized development of cross-border securities business, it will make corresponding adjustments to the services for existing investor accounts in China. Starting from June 12, 2026, Beijing time, for trading services within the country, new and additional trades for all varieties, including stocks, will be suspended, with only sell and close operations supported. For fund transfer services within the country, fund inflows will be suspended, while outflows will remain normal to ensure the safety of customer funds. Tiger stated that this adjustment will not affect the services provided to existing investors overseas, nor will it impact the safety of existing assets for all customers. Customers can continue to query their accounts, hold, and sell existing positions normally. (Securities Times)

Tiger International: Suspension of New and Additional Trades for Existing Investor Accounts in China Starting June 12

On June 2, NVIDIA announced that its Spectrum-X Ethernet silicon photonics technology has now entered full production. The next-generation Spectrum-X switch is built on Co-Packaged Optics (CPO) technology, supporting the NVIDIA Vera Rubin platform for horizontal scaling and cross-regional deployment of AI factories in data centers. The Spectrum-X Ethernet silicon photonics technology is a prime example of NVIDIA's full-stack collaborative design. Compared to networks using traditional transceivers, the Spectrum-X technology achieves a 5x improvement in energy efficiency, a 5x increase in AI uptime, and a 1.3x faster deployment time.

[NVIDIA Spectrum-X Ethernet Silicon Photonics Technology Now in Full Production, Achieving 5x Network Energy Efficiency Over Traditional Transceivers]On June 2, NVIDIA announced that its Spectrum-X Ethernet silicon photonics technology has now entered full production. The next-generation Spectrum-X switch is built on Co-Packaged Optics (CPO) technology, supporting the NVIDIA Vera Rubin platform for horizontal scaling and cross-regional deployment of AI factories in data centers. The Spectrum-X Ethernet silicon photonics technology is a prime example of NVIDIA's full-stack collaborative design. Compared to networks using traditional transceivers, the Spectrum-X technology achieves a 5x improvement in energy efficiency, a 5x increase in AI uptime, and a 1.3x faster deployment time.

NVIDIA Spectrum-X Ethernet Silicon Photonics Technology Now in Full Production, Achieving 5x Network Energy Efficiency Over Traditional Transceivers

On June 2, Iranian media reported that senior officials from the Iranian military and the Islamic Revolutionary Guard Corps stated that Iran has not yet utilized its full military capabilities and is prepared for various possible scenarios, maintaining the highest state of readiness. The Deputy Inspector General of the Central Headquarters of the Iranian Armed Forces, Asadi, indicated that Iran has not publicly demonstrated all its capabilities and possesses multiple response options that will be employed if necessary. He emphasized that the U.S. demand for Iran to 'completely surrender' is unattainable, asserting that 'the Iranian nation will never surrender.'

[Iranian Armed Forces Declare Readiness for Conflict at Highest Alert Level]On June 2, Iranian media reported that senior officials from the Iranian military and the Islamic Revolutionary Guard Corps stated that Iran has not yet utilized its full military capabilities and is prepared for various possible scenarios, maintaining the highest state of readiness. The Deputy Inspector General of the Central Headquarters of the Iranian Armed Forces, Asadi, indicated that Iran has not publicly demonstrated all its capabilities and possesses multiple response options that will be employed if necessary. He emphasized that the U.S. demand for Iran to 'completely surrender' is unattainable, asserting that 'the Iranian nation will never surrender.'

Iranian Armed Forces Declare Readiness for Conflict at Highest Alert Level

On June 2, according to Israel's i24News, a senior official from Israeli Prime Minister Netanyahu's team revealed that Axios's report regarding the late-night phone call between Netanyahu and Trump was inaccurate. Trump did not make personal comments regarding imprisonment nor did he claim that Netanyahu is hated globally. Instead, the tense focus of the call was on conflicting information on social media: Trump believed Netanyahu implied that the war was still fully ongoing, while Netanyahu thought Trump suggested a complete ceasefire. Trump did mention that defending Israel's global standing is very difficult and could provoke hatred. Ultimately, the call ended with an understanding: as long as Israel's homeland is not attacked, strikes on Beirut will be postponed.

[Israel Clarifies Reports on Trump-Netanyahu Phone Call]On June 2, according to Israel's i24News, a senior official from Israeli Prime Minister Netanyahu's team revealed that Axios's report regarding the late-night phone call between Netanyahu and Trump was inaccurate. Trump did not make personal comments regarding imprisonment nor did he claim that Netanyahu is hated globally. Instead, the tense focus of the call was on conflicting information on social media: Trump believed Netanyahu implied that the war was still fully ongoing, while Netanyahu thought Trump suggested a complete ceasefire. Trump did mention that defending Israel's global standing is very difficult and could provoke hatred. Ultimately, the call ended with an understanding: as long as Israel's homeland is not attacked, strikes on Beirut will be postponed.

Israel Clarifies Reports on Trump-Netanyahu Phone Call

On June 2, AI data center company Hyperscale (NYSE American: GPUS) announced that as of May 31, 2026, the company holds 704.3405 Bitcoins, valued at approximately $51.8 million based on a Bitcoin price of $73,579 on that date. The company did not purchase any Bitcoin on the open market last week.

[AI Data Center Company Hyperscale Holds Approximately $51.8 Million in BTC]On June 2, AI data center company Hyperscale (NYSE American: GPUS) announced that as of May 31, 2026, the company holds 704.3405 Bitcoins, valued at approximately $51.8 million based on a Bitcoin price of $73,579 on that date. The company did not purchase any Bitcoin on the open market last week.

AI Data Center Company Hyperscale Holds Approximately $51.8 Million in BTC

On June 2, the Shanghai Stock Exchange announced that Yushu Technology Co., Ltd.'s IPO review status for the Sci-Tech Innovation Board has changed to 'Registration Submitted'. This change indicates that the company has successfully passed the listing committee review of the Shanghai Stock Exchange and has entered the final issuance registration stage with the China Securities Regulatory Commission (CSRC), with only one last step remaining before official issuance and listing. The Shanghai Stock Exchange has completed the substantive review and submitted the materials to the CSRC, which typically completes the registration process within 20 working days (usually without substantive re-examination). Once registered, the company can initiate the pricing of its initial public offering and list on the Sci-Tech Innovation Board. This marks the sprint phase for Yushu Technology to become the first 'A-share embodiment intelligent robot stock', with the listing process proceeding very smoothly.

[Yushu Technology Moves One Step Closer to IPO, Status Changed to 'Registration Submitted']On June 2, the Shanghai Stock Exchange announced that Yushu Technology Co., Ltd.'s IPO review status for the Sci-Tech Innovation Board has changed to 'Registration Submitted'. This change indicates that the company has successfully passed the listing committee review of the Shanghai Stock Exchange and has entered the final issuance registration stage with the China Securities Regulatory Commission (CSRC), with only one last step remaining before official issuance and listing. The Shanghai Stock Exchange has completed the substantive review and submitted the materials to the CSRC, which typically completes the registration process within 20 working days (usually without substantive re-examination). Once registered, the company can initiate the pricing of its initial public offering and list on the Sci-Tech Innovation Board. This marks the sprint phase for Yushu Technology to become the first 'A-share embodiment intelligent robot stock', with the listing process proceeding very smoothly.

Yushu Technology Moves One Step Closer to IPO, Status Changed to 'Registration Submitted'

On June 2, AI data center company Hyperscale (NYSE American: GPUS) announced that as of May 31, 2026, the company holds 704.3405 bitcoins, valued at approximately $51.8 million based on a Bitcoin price of $73,579 on that date. The company did not purchase any Bitcoin on the open market last week.

[AI Data Center Company Hyperscale Holds Approximately $51.8 Million in BTC]On June 2, AI data center company Hyperscale (NYSE American: GPUS) announced that as of May 31, 2026, the company holds 704.3405 bitcoins, valued at approximately $51.8 million based on a Bitcoin price of $73,579 on that date. The company did not purchase any Bitcoin on the open market last week.

2 июня президент США Трамп заявил в интервью NBC News, что он не получил никаких новостей из Ирана относительно приостановки переговоров. Ранее Центральное командование США сообщило, что тысячи военнослужащих ВМС, ВВС и армии США оказывают поддержку текущим операциям блокады США против Ирана на море, в воздухе и на земле. По состоянию на 1 июня силы дали указание 121 коммерческому судну изменить курс и внедрили меры по приостановке на 5 кораблях для обеспечения бесперебойного выполнения связанных операций.

[Трамп: Нет новостей из Ирана о приостановке переговоров] 2 июня президент США Трамп заявил в интервью NBC News, что он не получил никаких новостей из Ирана относительно приостановки переговоров. Ранее Центральное командование США сообщило, что тысячи военнослужащих ВМС, ВВС и армии США оказывают поддержку текущим операциям блокады США против Ирана на море, в воздухе и на земле. По состоянию на 1 июня силы дали указание 121 коммерческому судну изменить курс и внедрили меры по приостановке на 5 кораблях для обеспечения бесперебойного выполнения связанных операций.

Трамп: Нет новостей из Ирана о приостановке переговоров

TechFlow Intelligence Report: Xiaomi Announces 200 Billion HKD Stock Buyback Plan, Spot Gold Falls Nearly 1%

CEO's Unexpected Passing: Will ONDO's 'Tokenization Narrative' Change?

4 Years After Terra’s Collapse, Hodlnaut’s Former CEO Faces Fraud Charges In Singapore Court

Hoskinson Says This Cardano App Could Become Crypto’s Most-Used By 2030

From Mining Pool to Mars: Why Is Wang Chun in the SpaceX Cockpit?

Retail Investors' 'Lead Brother' Serenity vs. Newly Minted Stock God Leopold: How Are the Two Top Hunters Mining AI's 'Physical Limits'?

One Article Deconstructs the Investment Methodology of 'Stock God Serenity'

Sharplink CEO: Ethereum's Future Is Playing Out Now

Kalshi and Coinbase Receive CFTC Approval, Ushering in the Most Regulation-Friendly Era for the Crypto Industry?

Weekly Token Unlocks: HYPE Unlocks $650M Worth of Tokens

Article

BTC drops below 72000 USDT

BTC drops below 71000 USDT

Трамп: Нет новостей из Ирана о приостановке переговоров

Disclaimer：The content above does not represent HTX's positions.，HTX does not provide any trading recommendations.。

All Comments0LatestHot

Trending Topics