7*24

Hot topic

Market

Activities

Spot silver has dropped 5.00% today, currently priced at $67.60 per ounce. New York silver futures have plummeted 7.00% today, currently at $67.55 per ounce.

[Spot Silver Drops 5.00% Today, Currently at $67.60/Ounce]Spot silver has dropped 5.00% today, currently priced at $67.60 per ounce. New York silver futures have plummeted 7.00% today, currently at $67.55 per ounce.

Spot Silver Drops 5.00% Today, Currently at $67.60/Ounce

According to monitoring by 1M AI News, Context Hub, an AI programming documentation service launched two weeks ago by DeepLearning.AI founder and Stanford University adjunct professor Wu Enda, has been exposed by security researchers for supply chain attack risks. Context Hub provides API documentation to programming agents via MCP servers, where contributors submit documentation through GitHub pull requests (PRs) and maintainers merge them for agents to read as needed. Mickey Shmueli, the creator of the alternative service lap.sh, released a proof of concept (PoC) attack, pointing out that 'there is no content review at any stage of this pipeline.' Shmueli created two sets of fake documentation targeting Plaid Link and Stripe Checkout, each embedding a forged PyPI package name, and tested them 40 times with three levels of Anthropic models: 1. Haiku wrote the malicious package into requirements.txt every time without showing any warnings in the output. 2. Sonnet issued warnings in 48% (19/40) of the tests but still wrote malicious dependencies in 53% (21/40) of the cases. 3. Opus performed the best, issuing warnings in 75% (30/40) of the tests and not writing malicious dependencies into the code. Attackers only need to submit a PR and have it merged to complete the poisoning, with a low review threshold: out of 97 closed PRs, 58 were merged. Shmueli noted that this is essentially a variant of indirect prompt injection, as AI models cannot reliably distinguish between data and instructions when processing content, and other community documentation services also lack adequate content review. Wu Enda did not respond to requests for comment.

[AI Programming Assistant Vulnerable to Supply Chain Attacks: Wu Enda's Context Hub Exposed]According to monitoring by 1M AI News, Context Hub, an AI programming documentation service launched two weeks ago by DeepLearning.AI founder and Stanford University adjunct professor Wu Enda, has been exposed by security researchers for supply chain attack risks. Context Hub provides API documentation to programming agents via MCP servers, where contributors submit documentation through GitHub pull requests (PRs) and maintainers merge them for agents to read as needed. Mickey Shmueli, the creator of the alternative service lap.sh, released a proof of concept (PoC) attack, pointing out that 'there is no content review at any stage of this pipeline.' Shmueli created two sets of fake documentation targeting Plaid Link and Stripe Checkout, each embedding a forged PyPI package name, and tested them 40 times with three levels of Anthropic models: 1. Haiku wrote the malicious package into requirements.txt every time without showing any warnings in the output. 2. Sonnet issued warnings in 48% (19/40) of the tests but still wrote malicious dependencies in 53% (21/40) of the cases. 3. Opus performed the best, issuing warnings in 75% (30/40) of the tests and not writing malicious dependencies into the code. Attackers only need to submit a PR and have it merged to complete the poisoning, with a low review threshold: out of 97 closed PRs, 58 were merged. Shmueli noted that this is essentially a variant of indirect prompt injection, as AI models cannot reliably distinguish between data and instructions when processing content, and other community documentation services also lack adequate content review. Wu Enda did not respond to requests for comment.

AI Programming Assistant Vulnerable to Supply Chain Attacks: Wu Enda's Context Hub Exposed

On March 26, according to The Wall Street Journal, Fannie Mae, a "super giant" in the U.S. housing mortgage market, will for the first time accept home loans backed by cryptocurrency assets.

[Fannie Mae to Accept Cryptocurrency-Backed Mortgages]On March 26, according to The Wall Street Journal, Fannie Mae, a "super giant" in the U.S. housing mortgage market, will for the first time accept home loans backed by cryptocurrency assets.

Fannie Mae to Accept Cryptocurrency-Backed Mortgages

On March 26, Hong Kong's Secretary for Financial Services and the Treasury, Christopher Hui, stated in response to questions from lawmakers that the People's Bank of China (PBOC) and the Hong Kong Monetary Authority (HKMA) are exploring arrangements and feasibility for upgrading digital currency wallets to increase usage limits, expand application scenarios, and enhance user experience. As the policies and technical details involved require further discussion, specific plans and timelines are yet to be finalized. Hui also disclosed the latest development data on digital renminbi in Hong Kong, noting that the local promotion of digital renminbi has seen significant progress, with a substantial increase in the number of wallet openings and merchant acceptance, marking the official entry of Hong Kong into the fast lane of digital payments.

[Accelerated Development of Digital Renminbi in Hong Kong: PBOC and HKMA Discuss Wallet Upgrade]On March 26, Hong Kong's Secretary for Financial Services and the Treasury, Christopher Hui, stated in response to questions from lawmakers that the People's Bank of China (PBOC) and the Hong Kong Monetary Authority (HKMA) are exploring arrangements and feasibility for upgrading digital currency wallets to increase usage limits, expand application scenarios, and enhance user experience. As the policies and technical details involved require further discussion, specific plans and timelines are yet to be finalized. Hui also disclosed the latest development data on digital renminbi in Hong Kong, noting that the local promotion of digital renminbi has seen significant progress, with a substantial increase in the number of wallet openings and merchant acceptance, marking the official entry of Hong Kong into the fast lane of digital payments.

Accelerated Development of Digital Renminbi in Hong Kong: PBOC and HKMA Discuss Wallet Upgrade

On March 26, Cointelegraph reported that Bitcoin has once again fallen below $70,000, with multiple on-chain and technical indicators suggesting that the current bear market is entering its later stage. In terms of on-chain data, Bitcoin's Net Unrealized Profit/Loss (NUPL) has dropped below 0.25, placing it in the 'hope/fear zone.' CryptoQuant analyst The Enigma Trader pointed out that this means approximately 40% of the circulating Bitcoin supply is in a loss state, compounded by the Fear and Greed Index dropping to a level of 15, indicating 'extreme fear,' which 'reflects pain and uncertainty.' He added that if NUPL rises back above 0.25, it would signal an entry into the optimistic zone, a transition that historically often coincides with enhanced price momentum. Glassnode noted that the 7-day moving average of relative unrealized losses has stabilized at 15%, stating, 'Historically, resolving this embedded loss level requires time, further price declines, or a sustained influx of fresh capital within a compressed timeframe.' Additionally, Bitcoin's entity-adjusted realized profit has plummeted from a peak of $3 billion per day in July 2025 to less than $100 million today, a decline of over 96%, which Glassnode describes as 'further evidence of demand exhaustion' and a 'textbook characteristic of the transition to the later stages of a bear market.' In terms of key price levels, Bitcoin has recently been oscillating within a range, with support at $64,000 and resistance at $72,000. Glassnode indicated that Bitcoin is struggling to maintain the 1-week to 1-month cost basis at $70,200, but the accumulation of buy orders at this level is insufficient, 'making the probability of breaking below this level non-negligible until more solid buying support is established.' The main support reference below is Bitcoin's realized price of about $54,000; the upper resistance is at the 1 to 3-month cost basis of $82,200, as well as a short-term holding zone above $84,000. Technical analyst CryptoPatel stated that Bitcoin's recent rise to $76,000 is merely a lower high, with higher time frame structures 'pointing lower,' and that the next truly noteworthy area is below $50,000.

[Analysis: Bitcoin Bear Market Enters Later Stage, $64,000 as Key Support Level]On March 26, Cointelegraph reported that Bitcoin has once again fallen below $70,000, with multiple on-chain and technical indicators suggesting that the current bear market is entering its later stage. In terms of on-chain data, Bitcoin's Net Unrealized Profit/Loss (NUPL) has dropped below 0.25, placing it in the 'hope/fear zone.' CryptoQuant analyst The Enigma Trader pointed out that this means approximately 40% of the circulating Bitcoin supply is in a loss state, compounded by the Fear and Greed Index dropping to a level of 15, indicating 'extreme fear,' which 'reflects pain and uncertainty.' He added that if NUPL rises back above 0.25, it would signal an entry into the optimistic zone, a transition that historically often coincides with enhanced price momentum. Glassnode noted that the 7-day moving average of relative unrealized losses has stabilized at 15%, stating, 'Historically, resolving this embedded loss level requires time, further price declines, or a sustained influx of fresh capital within a compressed timeframe.' Additionally, Bitcoin's entity-adjusted realized profit has plummeted from a peak of $3 billion per day in July 2025 to less than $100 million today, a decline of over 96%, which Glassnode describes as 'further evidence of demand exhaustion' and a 'textbook characteristic of the transition to the later stages of a bear market.' In terms of key price levels, Bitcoin has recently been oscillating within a range, with support at $64,000 and resistance at $72,000. Glassnode indicated that Bitcoin is struggling to maintain the 1-week to 1-month cost basis at $70,200, but the accumulation of buy orders at this level is insufficient, 'making the probability of breaking below this level non-negligible until more solid buying support is established.' The main support reference below is Bitcoin's realized price of about $54,000; the upper resistance is at the 1 to 3-month cost basis of $82,200, as well as a short-term holding zone above $84,000. Technical analyst CryptoPatel stated that Bitcoin's recent rise to $76,000 is merely a lower high, with higher time frame structures 'pointing lower,' and that the next truly noteworthy area is below $50,000.

Analysis: Bitcoin Bear Market Enters Later Stage, $64,000 as Key Support Level

On March 25th, Hong Kong's Secretary for Financial Services and the Treasury, Christopher Hui, stated in response to a legislator's inquiry that the People's Bank of China (PBOC) and the Hong Kong Monetary Authority (HKMA) are discussing arrangements and feasibility for upgrading the digital yuan wallet. The aim is to increase the usage limits, expand application scenarios, and enhance user experience. Specific plans and timelines are yet to be finalized as policy and technical details require further in-depth discussion. Hui also disclosed the latest development data for the digital yuan in Hong Kong, indicating that the promotion and adoption of the digital yuan locally have seen leapfrog development. Both the number of wallets opened and the scale of merchant acceptance have achieved substantial growth, officially propelling Hong Kong's digital payment era into high gear. (Yicai)

[Hong Kong Accelerates Digital Yuan Development; Central Bank and HKMA Discuss Wallet Upgrades]On March 25th, Hong Kong's Secretary for Financial Services and the Treasury, Christopher Hui, stated in response to a legislator's inquiry that the People's Bank of China (PBOC) and the Hong Kong Monetary Authority (HKMA) are discussing arrangements and feasibility for upgrading the digital yuan wallet. The aim is to increase the usage limits, expand application scenarios, and enhance user experience. Specific plans and timelines are yet to be finalized as policy and technical details require further in-depth discussion. Hui also disclosed the latest development data for the digital yuan in Hong Kong, indicating that the promotion and adoption of the digital yuan locally have seen leapfrog development. Both the number of wallets opened and the scale of merchant acceptance have achieved substantial growth, officially propelling Hong Kong's digital payment era into high gear. (Yicai)

Hong Kong Accelerates Digital Yuan Development; Central Bank and HKMA Discuss Wallet Upgrades

According to market news: The ByteDance team has open-sourced an AI team brain, DeerFlow 2.0, which has currently reached 47,000 stars on GitHub Trending. DeerFlow now natively supports connecting to multiple Agents based on the ACP protocol, including Codex, Claude Code, OpenClaw, and others.

[ByteDance Team Open-Sources AI Team Brain DeerFlow 2.0]According to market news: The ByteDance team has open-sourced an AI team brain, DeerFlow 2.0, which has currently reached 47,000 stars on GitHub Trending. DeerFlow now natively supports connecting to multiple Agents based on the ACP protocol, including Codex, Claude Code, OpenClaw, and others.

ByteDance Team Open-Sources AI Team Brain DeerFlow 2.0

On March 26, according to a Reuters survey: UK inflation is expected to drop to 2.8% in the next quarter and maintain an average level of 3.4% in the second half of 2026. Among 50 economists, 28 (or 56%) believe that the Bank of England will keep the benchmark interest rate at 3.75% until the end of the fourth quarter, a significant increase from just 7% in the survey conducted on March 12.

[Reuters Survey: UK Inflation Expected to Drop to 2.8% Next Quarter]On March 26, according to a Reuters survey: UK inflation is expected to drop to 2.8% in the next quarter and maintain an average level of 3.4% in the second half of 2026. Among 50 economists, 28 (or 56%) believe that the Bank of England will keep the benchmark interest rate at 3.75% until the end of the fourth quarter, a significant increase from just 7% in the survey conducted on March 12.

Reuters Survey: UK Inflation Expected to Drop to 2.8% Next Quarter

According to 1M AI News monitoring, Lin Junyang, former technical lead of Alibaba's Tongyi Qianwen, published a lengthy article on X, systematically elaborating on his judgment that the AI industry is shifting from 'reasoning thinking' to 'agentic thinking'. This is the first technical opinion piece he has publicly released since leaving the Qianwen team in early March. Lin Junyang believes that the core topic for the first half of 2025 is reasoning thinking, focusing on how to allocate more computing power during the reasoning phase, how to train with stronger reward signals, and how to control reasoning depth. However, the answer for the next stage is agentic thinking: models will no longer just 'think longer,' but will 'think to act,' continuously revising plans through interaction with the environment. In his article, he candidly reviewed the technical choices made by the Qianwen team. Qwen3 attempted to fuse thinking and instruction modes within the same model, supporting adjustable reasoning budgets. However, practical execution revealed significant differences in data distribution and behavioral objectives between the two modes: the instruction mode prioritizes conciseness, low latency, and format compliance, while the thinking mode aims to invest more tokens in difficult problems and maintain intermediate reasoning structures. If data curation for merging is not sufficiently refined, the result is often mediocrity in both. Consequently, the Qwen 2507 series ultimately chose to release separate Instruct and Thinking versions (available in 30B and 235B specifications) to allow for individual optimization. Anthropic took the opposite route, with Claude 3.7 Sonnet advocating from the outset that reasoning should be an integrated capability rather than a standalone model, allowing users to set their own thinking budgets. Lin Junyang proposed that the infrastructure for agent reinforcement learning is more challenging than traditional reasoning RL. Rollouts for reasoning RL are typically self-contained trajectories, manageable with static validators. Agent RL, on the other hand, requires models to embed a complete toolchain (browsers, terminals, sandboxes, APIs, memory systems), and training and inference must be decoupled, otherwise, rollout throughput will collapse. He elevated environment design to a status equivalent to model architecture, stating that 'environment construction is evolving from a side project to a true startup category.' He predicts that agentic thinking will become the mainstream mode of thought, potentially even replacing the excessively long, isolated internal monologues of traditional static reasoning. However, the biggest risk is reward hacking: once models gain access to real tools, they may learn to directly search for answers during RL training, utilize future information from repositories, or discover shortcuts to bypass tasks. The article concludes by pointing out that future competitive advantages will shift from better RL algorithms to better environment design, tighter integration of training and inference, and the system engineering capabilities of multi-agent collaboration.

[Former Qwen Lead Lin Junyang's First Post-Departure Article: AI Industry Shifts from 'Training Models' to 'Training Agents']According to 1M AI News monitoring, Lin Junyang, former technical lead of Alibaba's Tongyi Qianwen, published a lengthy article on X, systematically elaborating on his judgment that the AI industry is shifting from 'reasoning thinking' to 'agentic thinking'. This is the first technical opinion piece he has publicly released since leaving the Qianwen team in early March. Lin Junyang believes that the core topic for the first half of 2025 is reasoning thinking, focusing on how to allocate more computing power during the reasoning phase, how to train with stronger reward signals, and how to control reasoning depth. However, the answer for the next stage is agentic thinking: models will no longer just 'think longer,' but will 'think to act,' continuously revising plans through interaction with the environment. In his article, he candidly reviewed the technical choices made by the Qianwen team. Qwen3 attempted to fuse thinking and instruction modes within the same model, supporting adjustable reasoning budgets. However, practical execution revealed significant differences in data distribution and behavioral objectives between the two modes: the instruction mode prioritizes conciseness, low latency, and format compliance, while the thinking mode aims to invest more tokens in difficult problems and maintain intermediate reasoning structures. If data curation for merging is not sufficiently refined, the result is often mediocrity in both. Consequently, the Qwen 2507 series ultimately chose to release separate Instruct and Thinking versions (available in 30B and 235B specifications) to allow for individual optimization. Anthropic took the opposite route, with Claude 3.7 Sonnet advocating from the outset that reasoning should be an integrated capability rather than a standalone model, allowing users to set their own thinking budgets. Lin Junyang proposed that the infrastructure for agent reinforcement learning is more challenging than traditional reasoning RL. Rollouts for reasoning RL are typically self-contained trajectories, manageable with static validators. Agent RL, on the other hand, requires models to embed a complete toolchain (browsers, terminals, sandboxes, APIs, memory systems), and training and inference must be decoupled, otherwise, rollout throughput will collapse. He elevated environment design to a status equivalent to model architecture, stating that 'environment construction is evolving from a side project to a true startup category.' He predicts that agentic thinking will become the mainstream mode of thought, potentially even replacing the excessively long, isolated internal monologues of traditional static reasoning. However, the biggest risk is reward hacking: once models gain access to real tools, they may learn to directly search for answers during RL training, utilize future information from repositories, or discover shortcuts to bypass tasks. The article concludes by pointing out that future competitive advantages will shift from better RL algorithms to better environment design, tighter integration of training and inference, and the system engineering capabilities of multi-agent collaboration.

Former Qwen Lead Lin Junyang's First Post-Departure Article: AI Industry Shifts from 'Training Models' to 'Training Agents'

According to DailyHodl, macro expert and former Goldman Sachs executive Raoul Pal says that while negative sentiment across the crypto industry is at an all-time high, its fundamentals remain strong.
In a new interview with co-founder and host of Impact Theory Tom Bilyeu, Pal says investor negativity is higher than he has ever seen it, including during the Great Recession and the Dot.com bubble burst.
But Pal says the crypto space is bullish with widening adoption by institutional investors, noting that big tech is increasingly intertwining with the crypto industry.

Detail

【Macro guru Raoul Pal says crypto markets still fundamentally bullish despite peak negative sentiment】According to DailyHodl, macro expert and former Goldman Sachs executive Raoul Pal says that while negative sentiment across the crypto industry is at an all-time high, its fundamentals remain strong.
In a new interview with co-founder and host of Impact Theory Tom Bilyeu, Pal says investor negativity is higher than he has ever seen it, including during the Great Recession and the Dot.com bubble burst.
But Pal says the crypto space is bullish with widening adoption by institutional investors, noting that big tech is increasingly intertwining with the crypto industry.

Macro guru Raoul Pal says crypto markets still fundamentally bullish despite peak negative sentiment

Article

U.S. House Speaker: U.S. Military Will Not Engage in Ground Combat in Iran

Commodity ETFs See Record $11 Billion Outflow in March

House Speaker: 'Operation Epic Wrath' Nearing Completion

US House Speaker: US Military Will Not Conduct Ground Operations in Iran

cfg currently trades at 0.1475 USDT, up 14.08% in 24h, according to HTX market data.

Macro guru Raoul Pal says crypto markets still fundamentally bullish despite peak negative sentiment

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

Related Articles

All Comments0LatestHot

Trending Topics

Related Articles