7*24

Hot topic

Market

Activities

On March 31, the Hong Kong-flagged container ships 'COSCO North Arctic' and 'COSCO Indian Ocean', which had been stranded in the Persian Gulf for over a month, successfully passed through the Strait of Hormuz at 10:00 AM Beijing time on March 31. This marks the first passage of large Chinese vessels through the Strait of Hormuz since the end of February. According to vessel tracking data, the 'COSCO North Arctic' and 'COSCO Indian Ocean' set sail from waters near Dubai on the evening of March 30, navigating between Iran's largest islands, Qeshm Island and Larak Island, at a speed of approximately 10.4 knots, before passing through the narrow exit of the Strait of Hormuz and entering the waters of the Gulf of Oman. Vessel tracking data indicates that the two ships are expected to arrive at Port Klang in Malaysia on April 6. (Shenzhen News)

[Two Chinese Vessels Successfully Transit the Strait of Hormuz]On March 31, the Hong Kong-flagged container ships 'COSCO North Arctic' and 'COSCO Indian Ocean', which had been stranded in the Persian Gulf for over a month, successfully passed through the Strait of Hormuz at 10:00 AM Beijing time on March 31. This marks the first passage of large Chinese vessels through the Strait of Hormuz since the end of February. According to vessel tracking data, the 'COSCO North Arctic' and 'COSCO Indian Ocean' set sail from waters near Dubai on the evening of March 30, navigating between Iran's largest islands, Qeshm Island and Larak Island, at a speed of approximately 10.4 knots, before passing through the narrow exit of the Strait of Hormuz and entering the waters of the Gulf of Oman. Vessel tracking data indicates that the two ships are expected to arrive at Port Klang in Malaysia on April 6. (Shenzhen News)

Two Chinese Vessels Successfully Transit the Strait of Hormuz

On March 31, a strategist from UBS stated that the prolonged conflict in Iran and persistently high oil prices may indicate that the rise in gold prices is nearing its conclusion. Joni Teves, a precious metals strategist at the Swiss investment bank, mentioned in an interview last Friday that investors may currently be witnessing the tail end of the gold bull market. 'We believe that the gold price cycle generally aligns with the Federal Reserve's policy cycle, so we expect the upward trend to gradually fade by the end of the year, with gold prices consolidating at lower levels in the coming years,' she said. Teves added that UBS's baseline expectation is that after a period of consolidation, gold prices will reach new highs later this year as allocation funds continue to flow in. 'However, it should be noted that the ongoing conflict in the Middle East could lead to significant changes in the macroeconomic outlook and policy expectations, which would in turn affect the medium to long-term trend of gold prices.' (Jin Shi)

[UBS Warns: Gold Bull Market May Be Coming to an End]On March 31, a strategist from UBS stated that the prolonged conflict in Iran and persistently high oil prices may indicate that the rise in gold prices is nearing its conclusion. Joni Teves, a precious metals strategist at the Swiss investment bank, mentioned in an interview last Friday that investors may currently be witnessing the tail end of the gold bull market. 'We believe that the gold price cycle generally aligns with the Federal Reserve's policy cycle, so we expect the upward trend to gradually fade by the end of the year, with gold prices consolidating at lower levels in the coming years,' she said. Teves added that UBS's baseline expectation is that after a period of consolidation, gold prices will reach new highs later this year as allocation funds continue to flow in. 'However, it should be noted that the ongoing conflict in the Middle East could lead to significant changes in the macroeconomic outlook and policy expectations, which would in turn affect the medium to long-term trend of gold prices.' (Jin Shi)

UBS Warns: Gold Bull Market May Be Coming to an End

On March 31, according to Tencent Cloud, the Tencent WorkBuddy WeChat mini program was officially launched today. It supports dual modes of cloud sandbox and remote execution on local computers, allowing users to assign tasks directly via voice, photos, and file uploads on their mobile phones. The generated documents, PPTs, and videos can be downloaded with one click and easily forwarded.

[Tencent WorkBuddy Mini Program Launched with 'Cloud + Local' Dual Mode Operation]On March 31, according to Tencent Cloud, the Tencent WorkBuddy WeChat mini program was officially launched today. It supports dual modes of cloud sandbox and remote execution on local computers, allowing users to assign tasks directly via voice, photos, and file uploads on their mobile phones. The generated documents, PPTs, and videos can be downloaded with one click and easily forwarded.

Tencent WorkBuddy Mini Program Launched with 'Cloud + Local' Dual Mode Operation

On March 31, Google Research revealed that a quantum computer can crack Bitcoin private keys in just 9 minutes, which is 1 minute faster than the average Bitcoin block time. The threat of memory pool attacks has now become a real concern, prompting us to urgently transition to post-quantum cryptographic solutions.

[Google: Quantum Computer Can Crack Bitcoin Private Keys in Just 9 Minutes]On March 31, Google Research revealed that a quantum computer can crack Bitcoin private keys in just 9 minutes, which is 1 minute faster than the average Bitcoin block time. The threat of memory pool attacks has now become a real concern, prompting us to urgently transition to post-quantum cryptographic solutions.

Google: Quantum Computer Can Crack Bitcoin Private Keys in Just 9 Minutes

On March 31, the market structure experienced a subtle yet crucial shift as the Federal Reserve signaled a pause on interest rate hikes, leading the market to retract its rate increase expectations, indicating that policy pressure is no longer rising. Simultaneously, the Middle East conflict has been postponed to an April negotiation window, shifting risks from acute shocks to prolonged consumption. The combination of these factors has transitioned the market from 'rapid contraction' to 'dull oscillation under low liquidity.' This week, attention should still be paid to Wednesday's ISM and ADP reports, as well as Friday's non-farm payrolls; however, in an environment of decreased trading depth, data is more likely to trigger short-term volatility rather than provide a directional trend. Current market liquidity signals are clear: the bid-ask spread for U.S. Treasuries has widened by about 27%, and some market depths have dropped to around 10%, indicating that market makers are continuously contracting risk. Returning to BTC, the short-term structure has converged to a range of $66,100 to $68,500. The lower boundary of $66,100 serves as a repeatedly validated support zone, while the upper boundary of $68,500 represents the current main liquidation pressure. Movements within this range are fundamentally liquidity scans rather than trend-driven; only if it effectively surpasses $68,500 will there be an opportunity to unlock a higher liquidity space; otherwise, it will continue to experience repeated compression within the range.

[Bitunix Analyst: Policy Shift to Caution and Delayed War Push BTC into Low Liquidity Compression Between $66,100 and $68,500]On March 31, the market structure experienced a subtle yet crucial shift as the Federal Reserve signaled a pause on interest rate hikes, leading the market to retract its rate increase expectations, indicating that policy pressure is no longer rising. Simultaneously, the Middle East conflict has been postponed to an April negotiation window, shifting risks from acute shocks to prolonged consumption. The combination of these factors has transitioned the market from 'rapid contraction' to 'dull oscillation under low liquidity.' This week, attention should still be paid to Wednesday's ISM and ADP reports, as well as Friday's non-farm payrolls; however, in an environment of decreased trading depth, data is more likely to trigger short-term volatility rather than provide a directional trend. Current market liquidity signals are clear: the bid-ask spread for U.S. Treasuries has widened by about 27%, and some market depths have dropped to around 10%, indicating that market makers are continuously contracting risk. Returning to BTC, the short-term structure has converged to a range of $66,100 to $68,500. The lower boundary of $66,100 serves as a repeatedly validated support zone, while the upper boundary of $68,500 represents the current main liquidation pressure. Movements within this range are fundamentally liquidity scans rather than trend-driven; only if it effectively surpasses $68,500 will there be an opportunity to unlock a higher liquidity space; otherwise, it will continue to experience repeated compression within the range.

Bitunix Analyst: Policy Shift to Caution and Delayed War Push BTC into Low Liquidity Compression Between $66,100 and $68,500

On March 31, VIX index futures plummeted, at one point dropping over 4% to report at 27 points. According to the Wall Street Journal, Trump told aides he is willing to end the war without reopening the Strait of Hormuz. (Dongxin News Agency)

[VIX Index Futures Plunge Over 4%]On March 31, VIX index futures plummeted, at one point dropping over 4% to report at 27 points. According to the Wall Street Journal, Trump told aides he is willing to end the war without reopening the Strait of Hormuz. (Dongxin News Agency)

VIX Index Futures Plunge Over 4%

According to monitoring by 1M AI News, a developer reverse-engineered the 228MB binary of the standalone version of Claude Code using Ghidra, MITM proxy, and radare2, discovering two independent cache bugs that can increase API costs by 10-20 times without the user's knowledge. Related analysis has been submitted to GitHub (issue #40524) and marked by Anthropic as a regression bug for resolution. The first bug exists in the custom Bun runtime used by the standalone version. Each time an API request is made, the runtime looks for a billing identifier in the request body and replaces it, but the replacement logic hits the first matching item in the request body. If the conversation history happens to contain that string (for example, discussing Claude Code's internal billing mechanism), the replacement will hit the message content instead of the system prompt, causing a full cache rebuild for every request. A temporary workaround is to run it using npx @anthropic-ai/claude-code, as the npm package version does not include this replacement logic. The second bug affects all users who use --resume or --continue to restore sessions, introduced since v2.1.69. When restoring a session, the injection point for system supplementary information is different from that of a new session, leading to a complete mismatch of cache prefixes, resulting in the entire conversation history being rewritten from scratch instead of being read from the cache. Subsequent rounds return to normal, but the restore operation itself incurs significant additional overhead, and there is currently no external workaround. The developer estimates that for a long conversation of about 500,000 tokens, Bug 1 incurs an additional cost of approximately $0.04 per request, while Bug 2 incurs an additional cost of approximately $0.15 per restore, with the combined cost of a single request potentially exceeding $0.20. Previously, Anthropic engineer Lydia Hallie confirmed that users are reaching usage limits 'far faster than expected,' and several users in the Reddit comments believe that these two cache bugs may be one of the fundamental reasons for the abnormal consumption of usage.

[Reverse Engineering of Claude Code Reveals Two Cache Bugs, Potentially Increasing API Costs by 10-20 Times]According to monitoring by 1M AI News, a developer reverse-engineered the 228MB binary of the standalone version of Claude Code using Ghidra, MITM proxy, and radare2, discovering two independent cache bugs that can increase API costs by 10-20 times without the user's knowledge. Related analysis has been submitted to GitHub (issue #40524) and marked by Anthropic as a regression bug for resolution. The first bug exists in the custom Bun runtime used by the standalone version. Each time an API request is made, the runtime looks for a billing identifier in the request body and replaces it, but the replacement logic hits the first matching item in the request body. If the conversation history happens to contain that string (for example, discussing Claude Code's internal billing mechanism), the replacement will hit the message content instead of the system prompt, causing a full cache rebuild for every request. A temporary workaround is to run it using npx @anthropic-ai/claude-code, as the npm package version does not include this replacement logic. The second bug affects all users who use --resume or --continue to restore sessions, introduced since v2.1.69. When restoring a session, the injection point for system supplementary information is different from that of a new session, leading to a complete mismatch of cache prefixes, resulting in the entire conversation history being rewritten from scratch instead of being read from the cache. Subsequent rounds return to normal, but the restore operation itself incurs significant additional overhead, and there is currently no external workaround. The developer estimates that for a long conversation of about 500,000 tokens, Bug 1 incurs an additional cost of approximately $0.04 per request, while Bug 2 incurs an additional cost of approximately $0.15 per restore, with the combined cost of a single request potentially exceeding $0.20. Previously, Anthropic engineer Lydia Hallie confirmed that users are reaching usage limits 'far faster than expected,' and several users in the Reddit comments believe that these two cache bugs may be one of the fundamental reasons for the abnormal consumption of usage.

Reverse Engineering of Claude Code Reveals Two Cache Bugs, Potentially Increasing API Costs by 10-20 Times

Musk stated on X that the future of AI primarily lies in video understanding and generation, "because photons are the highest bandwidth communication form to date, these are key tools on the path to AGI," and noted that xAI's image generation tool Grok Imagine is "profitable, not a loss-making project." This statement comes in the context of OpenAI's announcement last week to shut down its AI video platform Sora. According to reports from WSJ and several media outlets, Sora has generated approximately $2.1 million in revenue since its launch in September last year, with an average daily operating loss reported at around $1 million. OpenAI also terminated its $1 billion partnership with Disney. Musk did not disclose specific revenue figures for Grok Imagine. He previously stated on March 25 that "the next version of Grok Imagine will be outstanding, and we are doubling our investment," and described it as "a tool more important than people realize."

[Musk Claims Grok Imagine is Profitable for xAI, Video is Key Tool for AGI]Musk stated on X that the future of AI primarily lies in video understanding and generation, "because photons are the highest bandwidth communication form to date, these are key tools on the path to AGI," and noted that xAI's image generation tool Grok Imagine is "profitable, not a loss-making project." This statement comes in the context of OpenAI's announcement last week to shut down its AI video platform Sora. According to reports from WSJ and several media outlets, Sora has generated approximately $2.1 million in revenue since its launch in September last year, with an average daily operating loss reported at around $1 million. OpenAI also terminated its $1 billion partnership with Disney. Musk did not disclose specific revenue figures for Grok Imagine. He previously stated on March 25 that "the next version of Grok Imagine will be outstanding, and we are doubling our investment," and described it as "a tool more important than people realize."

Musk Claims Grok Imagine is Profitable for xAI, Video is Key Tool for AGI

According to monitoring by 1M AI News, Xiaomi's MiClaw team has launched a system input method that contains serious security oversights. Users discovered that by repeatedly clicking on the version number of the input method, a debugging page could be accessed, which directly exposed the API call addresses, API Key, model provider, and model name of the AI service in plain text. The leaked API address points to ByteDance's cloud service platform, Volcano Engine's Ark interface, using the doubao-seed-1-6-lite-251015 model from the Doubao series. Based on the prompts, this AI feature is used for post-processing voice input, responsible for correcting typos, grammatical errors, and adding punctuation in the transcribed text. Users confirmed that the key is real and valid, allowing direct calls from external platforms; Xiaomi has reportedly changed the key since then. Decompiling the code also revealed engineering quality issues: developers used an if ("fixed string".length() > 0) method to check whether a hardcoded string that is always true is non-empty, a practice that would not appear in any normal code review process. Additionally, Xiaomi was found to have hardcoded the API key of the AI company Moonshot in the code submissions of its open-source project mone on GitHub, with the submission date being January 2025, and no changes recorded since.

[Xiaomi's New Input Method Exposes AI Model Keys Due to Security Flaw]According to monitoring by 1M AI News, Xiaomi's MiClaw team has launched a system input method that contains serious security oversights. Users discovered that by repeatedly clicking on the version number of the input method, a debugging page could be accessed, which directly exposed the API call addresses, API Key, model provider, and model name of the AI service in plain text. The leaked API address points to ByteDance's cloud service platform, Volcano Engine's Ark interface, using the doubao-seed-1-6-lite-251015 model from the Doubao series. Based on the prompts, this AI feature is used for post-processing voice input, responsible for correcting typos, grammatical errors, and adding punctuation in the transcribed text. Users confirmed that the key is real and valid, allowing direct calls from external platforms; Xiaomi has reportedly changed the key since then. Decompiling the code also revealed engineering quality issues: developers used an if ("fixed string".length() > 0) method to check whether a hardcoded string that is always true is non-empty, a practice that would not appear in any normal code review process. Additionally, Xiaomi was found to have hardcoded the API key of the AI company Moonshot in the code submissions of its open-source project mone on GitHub, with the submission date being January 2025, and no changes recorded since.

Xiaomi's New Input Method Exposes AI Model Keys Due to Security Flaw

According to Forkast, Bank of Thailand official Vachira Arromdee said on Tuesday the four countries participating in the multinational central bank digital currency (CBDC) have agreed to end the use case of paying foreign CBDC domestically.
Paying foreign currency through mBridge domestically can have “crazy substitution effects” on the monetary policy, said the Executive Advisor of Bank of Thailand’s CBDC Projects at a panel of the Hong Kong Fintech Week.
The mBridge, or mCBDC, is a collaborative CBDC project between the HKMA and the central banks of Thailand, China, the United Arab Emirates and the BIS to facilitate multi-currency cross-border payments.

Detail

【mBridge participants to end domestic payments of foreign CBDC】According to Forkast, Bank of Thailand official Vachira Arromdee said on Tuesday the four countries participating in the multinational central bank digital currency (CBDC) have agreed to end the use case of paying foreign CBDC domestically.
Paying foreign currency through mBridge domestically can have “crazy substitution effects” on the monetary policy, said the Executive Advisor of Bank of Thailand’s CBDC Projects at a panel of the Hong Kong Fintech Week.
The mBridge, or mCBDC, is a collaborative CBDC project between the HKMA and the central banks of Thailand, China, the United Arab Emirates and the BIS to facilitate multi-currency cross-border payments.

mBridge participants to end domestic payments of foreign CBDC

Article

Grok on X Begins Automatic Translation and Recommendation of Foreign Language Posts

Strategy Accumulates 45,000 BTC in Last 30 Days, Fastest Growth in a Year

Reuters Opinion: Iran May Cause Widespread Damage to Gulf Energy Infrastructure if US Launches Ground Offensive

Predicted Market Transactions Reach 191 Million in March

Trump's Influence Fades: Market Immune to Verbal Assurances as U.S. Stocks Drop for Five Consecutive Weeks

mBridge participants to end domestic payments of foreign CBDC

Descargo de responsabilidad：El contenido anterior no representa las posiciones de HTX.，HTX no ofrece ninguna recomendación de trading.。

Artículos relacionados

Todos los comentarios0Lo más recientePopular

Temas populares

Artículos relacionados