7*24

Hot topic

Market

Activities

According to monitoring by 1M AI News, OpenAI, Anthropic, and Google are sharing information through the industry nonprofit organization Frontier Model Forum to jointly detect violations of their terms of service related to 'adversarial distillation.' This organization was jointly established by the three companies along with Microsoft in 2023. Distillation is a technique that uses the outputs of existing models to train new models, replicating the former's capabilities at a lower cost. AI companies often use distillation to create smaller and more efficient versions of their models, but unauthorized third-party distillation is controversial in the industry. OpenAI confirmed its participation in this collaboration and cited a memo it previously submitted to the U.S. Congress, stating that deep learning attempts to 'ride on the coattails of the capabilities developed by OpenAI and other leading U.S. labs.' Google, Anthropic, and Frontier Model Forum declined to comment. U.S. government officials estimate that unauthorized distillation results in billions of dollars in profit losses for Silicon Valley labs each year. This collaboration draws on practices in the cybersecurity industry where companies share attack intelligence. The AI action plan released by the Trump administration last year also called for the establishment of dedicated information sharing and analysis centers. However, due to uncertainties regarding the boundaries of what can be shared under current antitrust laws, the scope of information sharing remains limited. The three companies have not provided public evidence indicating the extent to which Chinese model innovations rely on distillation, but they claim that the frequency of such behavior can be measured by the number of large-scale data requests.

[OpenAI, Anthropic, and Google Collaborate to Share Intelligence Against China's AI Model Distillation]According to monitoring by 1M AI News, OpenAI, Anthropic, and Google are sharing information through the industry nonprofit organization Frontier Model Forum to jointly detect violations of their terms of service related to 'adversarial distillation.' This organization was jointly established by the three companies along with Microsoft in 2023. Distillation is a technique that uses the outputs of existing models to train new models, replicating the former's capabilities at a lower cost. AI companies often use distillation to create smaller and more efficient versions of their models, but unauthorized third-party distillation is controversial in the industry. OpenAI confirmed its participation in this collaboration and cited a memo it previously submitted to the U.S. Congress, stating that deep learning attempts to 'ride on the coattails of the capabilities developed by OpenAI and other leading U.S. labs.' Google, Anthropic, and Frontier Model Forum declined to comment. U.S. government officials estimate that unauthorized distillation results in billions of dollars in profit losses for Silicon Valley labs each year. This collaboration draws on practices in the cybersecurity industry where companies share attack intelligence. The AI action plan released by the Trump administration last year also called for the establishment of dedicated information sharing and analysis centers. However, due to uncertainties regarding the boundaries of what can be shared under current antitrust laws, the scope of information sharing remains limited. The three companies have not provided public evidence indicating the extent to which Chinese model innovations rely on distillation, but they claim that the frequency of such behavior can be measured by the number of large-scale data requests.

OpenAI, Anthropic, and Google Collaborate to Share Intelligence Against China's AI Model Distillation

cfg Surges 10.30% in 24H, Currently at 0.1757 USDT

【cfg currently trades at 0.1757 USDT, up 10.30% in 24h, according to HTX market data.】cfg Surges 10.30% in 24H, Currently at 0.1757 USDT

cfg currently trades at 0.1757 USDT, up 10.30% in 24h, according to HTX market data.

According to monitoring by 1M AI News, OpenAI has announced the opening of applications for the Safety Fellowship, a pilot program aimed at external researchers, engineers, and practitioners. The fellowship will last approximately five months, from September 14, 2026, to February 5, 2027, focusing on AI safety and alignment research. Priority areas include safety assessment, ethics, robustness, scalable mitigation strategies, privacy-preserving safety methods, agent supervision, and high-risk abuse scenarios. Selected fellows will receive a monthly stipend, compute support, and mentorship from OpenAI, with the option to work at the Constellation co-working space in Berkeley or remotely. At the end of the program, participants are required to produce substantial research outputs such as papers, benchmark datasets, or data sets. Scholars will receive API credits and related resources but will not have access to OpenAI's internal systems. Applications are open to candidates from multidisciplinary backgrounds including computer science, social sciences, cybersecurity, privacy, and human-computer interaction, and require a letter of recommendation. The application deadline is May 3, with results to be announced by July 25.

[OpenAI Launches Safety Fellowship for External Researchers with Compute Power and Stipends]According to monitoring by 1M AI News, OpenAI has announced the opening of applications for the Safety Fellowship, a pilot program aimed at external researchers, engineers, and practitioners. The fellowship will last approximately five months, from September 14, 2026, to February 5, 2027, focusing on AI safety and alignment research. Priority areas include safety assessment, ethics, robustness, scalable mitigation strategies, privacy-preserving safety methods, agent supervision, and high-risk abuse scenarios. Selected fellows will receive a monthly stipend, compute support, and mentorship from OpenAI, with the option to work at the Constellation co-working space in Berkeley or remotely. At the end of the program, participants are required to produce substantial research outputs such as papers, benchmark datasets, or data sets. Scholars will receive API credits and related resources but will not have access to OpenAI's internal systems. Applications are open to candidates from multidisciplinary backgrounds including computer science, social sciences, cybersecurity, privacy, and human-computer interaction, and require a letter of recommendation. The application deadline is May 3, with results to be announced by July 25.

OpenAI Launches Safety Fellowship for External Researchers with Compute Power and Stipends

According to monitoring by 1M AI News, AI programming tool Cursor has released a technical blog introducing its self-developed MoE (Mixture of Experts) inference acceleration method, Warp Decode. This method targets small-batch token generation scenarios on NVIDIA's Blackwell GPU, flipping the traditional expert-centric parallel strategy to an output-centric approach: each warp (the smallest scheduling unit composed of 32 parallel processing units) in the GPU is responsible for calculating a single output value, independently traversing all routed experts and completing accumulation in registers without any cross-warp synchronization or intermediate buffers. The traditional MoE inference pipeline consists of 8 stages, 5 of which are solely for moving data for expert views without performing actual computations. Warp Decode compresses the entire MoE computation layer into 2 CUDA kernels, eliminating intermediate steps such as padding, scattering, and merging, reducing over 32KB of intermediate buffer read/write per token. Tested on NVIDIA's B200 GPU with a Qwen-3 style model, Warp Decode achieved a 1.84x end-to-end decoding throughput improvement, and due to computing entirely in BF16/FP32 precision, it avoided intermediate quantization loss, resulting in an output accuracy that is 1.4 times closer to the FP32 benchmark compared to traditional paths. In terms of hardware bandwidth utilization, with a batch size of 32, it sustained a throughput of 3.95 TB/s, approximately 58% of the B200's peak bandwidth (6.8 TB/s). This optimization directly accelerates the development iteration and version release pace of Cursor's self-developed programming model, Composer.

[Cursor Unveils MoE Inference Optimization Technology Warp Decode, Achieving 1.84x Throughput Improvement on Blackwell GPU]According to monitoring by 1M AI News, AI programming tool Cursor has released a technical blog introducing its self-developed MoE (Mixture of Experts) inference acceleration method, Warp Decode. This method targets small-batch token generation scenarios on NVIDIA's Blackwell GPU, flipping the traditional expert-centric parallel strategy to an output-centric approach: each warp (the smallest scheduling unit composed of 32 parallel processing units) in the GPU is responsible for calculating a single output value, independently traversing all routed experts and completing accumulation in registers without any cross-warp synchronization or intermediate buffers. The traditional MoE inference pipeline consists of 8 stages, 5 of which are solely for moving data for expert views without performing actual computations. Warp Decode compresses the entire MoE computation layer into 2 CUDA kernels, eliminating intermediate steps such as padding, scattering, and merging, reducing over 32KB of intermediate buffer read/write per token. Tested on NVIDIA's B200 GPU with a Qwen-3 style model, Warp Decode achieved a 1.84x end-to-end decoding throughput improvement, and due to computing entirely in BF16/FP32 precision, it avoided intermediate quantization loss, resulting in an output accuracy that is 1.4 times closer to the FP32 benchmark compared to traditional paths. In terms of hardware bandwidth utilization, with a batch size of 32, it sustained a throughput of 3.95 TB/s, approximately 58% of the B200's peak bandwidth (6.8 TB/s). This optimization directly accelerates the development iteration and version release pace of Cursor's self-developed programming model, Composer.

Cursor Unveils MoE Inference Optimization Technology Warp Decode, Achieving 1.84x Throughput Improvement on Blackwell GPU

On April 7, the contract whale 'Set 10 Major Goals' (@Jason60704294) posted a limit order for a long position on X at 1 AM this morning. The screenshot shows a limit price of $69,574.7, with an increase of 270.37 Bitcoin in long positions. Bitcoin fell below this purchase price this morning, so it is expected that this order has been executed. Yesterday afternoon, Bitcoin briefly surpassed the $70,000 mark, and the BTC long position of whale 'Set 10 Major Goals' once showed an unrealized profit of $5.52 million.

[Whale 'Set 10 Major Goals' Increases Bitcoin Long Position]On April 7, the contract whale 'Set 10 Major Goals' (@Jason60704294) posted a limit order for a long position on X at 1 AM this morning. The screenshot shows a limit price of $69,574.7, with an increase of 270.37 Bitcoin in long positions. Bitcoin fell below this purchase price this morning, so it is expected that this order has been executed. Yesterday afternoon, Bitcoin briefly surpassed the $70,000 mark, and the BTC long position of whale 'Set 10 Major Goals' once showed an unrealized profit of $5.52 million.

Whale 'Set 10 Major Goals' Increases Bitcoin Long Position

On April 7, CryptoQuant reported that the Bitcoin network activity index has rebounded after months of decline, with daily transaction volume reaching approximately 615,000, the highest level since November 2024. This includes various metrics such as active addresses, transactions, and UTXOs. The analysis also pointed out that current Bitcoin fees remain low, suggesting that exchanges and large holders may be engaging in UTXO consolidation and fund restructuring. The surge in activity is partly attributed to operational behavior rather than purely organic demand.

[Bitcoin Network Activity Rebounds, Daily Transaction Volume Hits Highest Level Since November 2024]On April 7, CryptoQuant reported that the Bitcoin network activity index has rebounded after months of decline, with daily transaction volume reaching approximately 615,000, the highest level since November 2024. This includes various metrics such as active addresses, transactions, and UTXOs. The analysis also pointed out that current Bitcoin fees remain low, suggesting that exchanges and large holders may be engaging in UTXO consolidation and fund restructuring. The surge in activity is partly attributed to operational behavior rather than purely organic demand.

Bitcoin Network Activity Rebounds, Daily Transaction Volume Hits Highest Level Since November 2024

On April 7, Ethena officially announced that after receiving approval from its independent risk committee, USDe will expand its collateral types. Ethena stated that this diversification of collateral assets includes four directions, each a natural extension of Ethena's existing business: 1. Over-collateralized institutional stablecoin lending: Custody of collateral assets provided by borrowers through third-party institutional custodians; 2. More real-world assets (RWA) beyond U.S. Treasury Bills (T-Bills): Expanding to more high-quality, liquid credit assets; 3. Stock and commodity basis trading: Extending Ethena's delta-neutral hedging methods, previously applied to crypto assets, to traditional assets; 4. Prime lending for trading institutions: Providing funding to trading firms through over-collateralized loans, with counterparties assuming the risk of the trading platform.

[Ethena Expands Neutral Arbitrage Trading to Stock and Commodity Markets]On April 7, Ethena officially announced that after receiving approval from its independent risk committee, USDe will expand its collateral types. Ethena stated that this diversification of collateral assets includes four directions, each a natural extension of Ethena's existing business: 1. Over-collateralized institutional stablecoin lending: Custody of collateral assets provided by borrowers through third-party institutional custodians; 2. More real-world assets (RWA) beyond U.S. Treasury Bills (T-Bills): Expanding to more high-quality, liquid credit assets; 3. Stock and commodity basis trading: Extending Ethena's delta-neutral hedging methods, previously applied to crypto assets, to traditional assets; 4. Prime lending for trading institutions: Providing funding to trading firms through over-collateralized loans, with counterparties assuming the risk of the trading platform.

Ethena Expands Neutral Arbitrage Trading to Stock and Commodity Markets

On April 7, according to monitoring by Lookonchain, crypto mining company MARA transferred 250 BTC (worth approximately $17.37 million) again three hours ago. Previously, from March 4 to March 25, MARA sold 15,133 BTC (worth $1.1 billion).

[Crypto Mining Company MARA Transfers 250 BTC Again]On April 7, according to monitoring by Lookonchain, crypto mining company MARA transferred 250 BTC (worth approximately $17.37 million) again three hours ago. Previously, from March 4 to March 25, MARA sold 15,133 BTC (worth $1.1 billion).

Crypto Mining Company MARA Transfers 250 BTC Again

On April 7, according to on-chain analyst Yu Jin, Spark transferred 414,000 USDS to the buyback address 7 hours ago, marking the start of the second month of SPK buybacks. In the first month, 572,000 USDS were used to repurchase 26.66 million SPK. The buyback funds of 414,000 USDS for the second month have now been transferred to the buyback address and the buyback has commenced.

[Spark Transfers 414,000 USDS to Buyback Address, Initiates Second Month of SPK Buyback]On April 7, according to on-chain analyst Yu Jin, Spark transferred 414,000 USDS to the buyback address 7 hours ago, marking the start of the second month of SPK buybacks. In the first month, 572,000 USDS were used to repurchase 26.66 million SPK. The buyback funds of 414,000 USDS for the second month have now been transferred to the buyback address and the buyback has commenced.

Spark Transfers 414,000 USDS to Buyback Address, Initiates Second Month of SPK Buyback

The data shows that the premium of the Grayscale Bitcoin Trust (GBTC) reached -45.2%, a new record low. In addition, the ETH trust premium is -44.65%, the BCH trust premium is -32.61%, the LTC trust premium is -48.28%, and the ETC trust premium is -66.99%.

Detail

【The Grayscale GBTC premium rate reached -45.2%, a new record low】The data shows that the premium of the Grayscale Bitcoin Trust (GBTC) reached -45.2%, a new record low. In addition, the ETH trust premium is -44.65%, the BCH trust premium is -32.61%, the LTC trust premium is -48.28%, and the ETC trust premium is -66.99%.

The Grayscale GBTC premium rate reached -45.2%, a new record low

Article

Arc Blockchain Releases Post-Quantum Cryptography Roadmap

China's AI Model Usage Surges Over 31%, Outpacing the U.S. for Five Consecutive Weeks

Software Engineering Job Openings Reach Three-Year High; a16z Co-Founder Andreessen Calls AI Unemployment Claims 'Completely False'

Iranian Senior Official: Iran Will Not Accept 'Diktat'

BTC broke through 70000  USDT

The Grayscale GBTC premium rate reached -45.2%, a new record low

Застереження：Наведені вище матеріали не представляють позицію компанії HTX.，HTX не дає жодних торгових рекомендацій.。

Схожі статті

Усі коментарі0НовіПопулярно

Популярні теми

Схожі статті