7*24

Hot topic

Market

Activities

On May 9, the Solana ecosystem project Solstice opened the inquiry for its first quarter airdrop, allowing users to check their SLX token airdrop shares on the official Earn Flares page. Previously, Solstice had opened registration for the first quarter airdrop from April 14 to April 21, with a registration fee of 0.075 SOL, and the window was only open for 7 days. Only registered users are eligible to claim the first quarter SLX airdrop; if not registered, the first quarter points and corresponding SLX airdrop shares will be reclaimed.

[Solstice Opens First Quarter Airdrop Inquiry; Unregistered Users' Points and Airdrops Reclaimed]On May 9, the Solana ecosystem project Solstice opened the inquiry for its first quarter airdrop, allowing users to check their SLX token airdrop shares on the official Earn Flares page. Previously, Solstice had opened registration for the first quarter airdrop from April 14 to April 21, with a registration fee of 0.075 SOL, and the window was only open for 7 days. Only registered users are eligible to claim the first quarter SLX airdrop; if not registered, the first quarter points and corresponding SLX airdrop shares will be reclaimed.

Solstice Opens First Quarter Airdrop Inquiry; Unregistered Users' Points and Airdrops Reclaimed

According to monitoring by Dongcha Beating, MiniMax released a technical blog disclosing the root cause investigation of its M2 series large model's inability to output the name 'Ma Jiaqi'. The investigation started from a specific case and ultimately revealed a systemic degradation issue affecting the entire vocabulary. The root cause was identified as the tokenizer (a component that segments text into units for model processing) merging 'Jiaqi' into a standalone token during training. In the pre-training phase, the model encountered a large amount of internet text and learned this token; however, in the post-training dialogue data, there were fewer than 5 samples containing 'Jiaqi'. During post-training, high-frequency tokens like tool_call markers and code symbols continuously updated the surrounding vector space, pushing low-frequency tokens like 'Jiaqi' in the wrong direction. The model still 'recognizes' Ma Jiaqi and can accurately respond with related information; it has merely lost the ability to output this token. The team subsequently conducted a comprehensive scan of approximately 200,000 tokens in the complete vocabulary and found that about 4.9% of tokens had significantly degraded. The most severe degradation was observed in Japanese: 29.7% of Japanese tokens showed significant degradation, far exceeding Korean (3.3%), Russian (3.7%), Chinese (3.9%), and English (3.5%). Other notably degraded tokens included internet SEO garbage terms like 'legendary private server' and 'painless abortion', with mechanisms identical to that of 'Jiaqi'. The severe degradation in Japanese also solved an old mystery. Previously, the model occasionally mixed in Russian or Korean characters in Japanese dialogues, but the cause was unknown. This analysis indicated that after the parameter drift of Japanese tokens, they became confused with tokens from other languages in the vector space, leading to incorrect activation of Japanese tokens (language mixing) and pushing adjacent low-frequency Chinese tokens out of the normal probability range (token forgetting). The solution is to construct a synthetic dataset covering the entire vocabulary, allowing the model to practice each token through simple repetition tasks. The results were immediate: the proportion of Russian characters mixed into Japanese responses dropped from 47% to 1%, and the stability of output parameters for the entire vocabulary (cosine similarity) increased from a low of 0.329 to all above 0.97.

[Why Large Models Fail to Generate 'Ma Jiaqi': MiniMax's Token Analysis Reveals Nearly 5% of Tokens Forgotten in Post-Training]According to monitoring by Dongcha Beating, MiniMax released a technical blog disclosing the root cause investigation of its M2 series large model's inability to output the name 'Ma Jiaqi'. The investigation started from a specific case and ultimately revealed a systemic degradation issue affecting the entire vocabulary. The root cause was identified as the tokenizer (a component that segments text into units for model processing) merging 'Jiaqi' into a standalone token during training. In the pre-training phase, the model encountered a large amount of internet text and learned this token; however, in the post-training dialogue data, there were fewer than 5 samples containing 'Jiaqi'. During post-training, high-frequency tokens like tool_call markers and code symbols continuously updated the surrounding vector space, pushing low-frequency tokens like 'Jiaqi' in the wrong direction. The model still 'recognizes' Ma Jiaqi and can accurately respond with related information; it has merely lost the ability to output this token. The team subsequently conducted a comprehensive scan of approximately 200,000 tokens in the complete vocabulary and found that about 4.9% of tokens had significantly degraded. The most severe degradation was observed in Japanese: 29.7% of Japanese tokens showed significant degradation, far exceeding Korean (3.3%), Russian (3.7%), Chinese (3.9%), and English (3.5%). Other notably degraded tokens included internet SEO garbage terms like 'legendary private server' and 'painless abortion', with mechanisms identical to that of 'Jiaqi'. The severe degradation in Japanese also solved an old mystery. Previously, the model occasionally mixed in Russian or Korean characters in Japanese dialogues, but the cause was unknown. This analysis indicated that after the parameter drift of Japanese tokens, they became confused with tokens from other languages in the vector space, leading to incorrect activation of Japanese tokens (language mixing) and pushing adjacent low-frequency Chinese tokens out of the normal probability range (token forgetting). The solution is to construct a synthetic dataset covering the entire vocabulary, allowing the model to practice each token through simple repetition tasks. The results were immediate: the proportion of Russian characters mixed into Japanese responses dropped from 47% to 1%, and the stability of output parameters for the entire vocabulary (cosine similarity) increased from a low of 0.329 to all above 0.97.

Why Large Models Fail to Generate 'Ma Jiaqi': MiniMax's Token Analysis Reveals Nearly 5% of Tokens Forgotten in Post-Training

PAYAI fluctuated by 76.29% within 24 hours, current price $0.007787

【PAYAI fluctuated over 50% in 24 hours!】PAYAI fluctuated by 76.29% within 24 hours, current price $0.007787

PAYAI fluctuated over 50% in 24 hours!

On May 9, it was reported that as the cost of memory chips rises and ByteDance accelerates its investment in artificial intelligence (AI), the company has increased its AI infrastructure spending plan for this year by 25% to 200 billion yuan.

[ByteDance Plans to Increase AI Infrastructure Spending by 25%]On May 9, it was reported that as the cost of memory chips rises and ByteDance accelerates its investment in artificial intelligence (AI), the company has increased its AI infrastructure spending plan for this year by 25% to 200 billion yuan.

ByteDance Plans to Increase AI Infrastructure Spending by 25%

On May 9, Bitcoin mining company Bitdeer, listed on NASDAQ, released its latest Bitcoin holdings data. As of the week ending May 8, its Bitcoin mining output was 193.8 BTC, but it also sold 193.8 BTC during the same period, resulting in a net addition of 0 BTC. Bitdeer currently maintains zero Bitcoin holdings.

[Bitdeer Maintains Zero Bitcoin Holdings, Sells 193.8 BTC This Week]On May 9, Bitcoin mining company Bitdeer, listed on NASDAQ, released its latest Bitcoin holdings data. As of the week ending May 8, its Bitcoin mining output was 193.8 BTC, but it also sold 193.8 BTC during the same period, resulting in a net addition of 0 BTC. Bitdeer currently maintains zero Bitcoin holdings.

Bitdeer Maintains Zero Bitcoin Holdings, Sells 193.8 BTC This Week

On May 9, Trump Media & Technology Group (TMTG) released its financial report for Q1 2026, showing a net loss of $405.9 million, with unrealized losses from digital assets and equity securities amounting to $368.7 million, making up the majority of the total loss. According to CoinGecko data, TMTG's crypto asset portfolio is currently valued at $821.9 million, with a cost basis of $1.24 billion, resulting in an overall unrealized loss of approximately $423 million. The asset portfolio includes 9,542 BTC (valued at $767 million, with an average price of $118,529) and 756 million CRO (valued at $54 million). Bitcoin fell about 22% in the quarter, marking the worst quarterly performance since 2018. On the operational side, Truth Social generated only $900,000 in revenue, but achieved an operating cash flow of $17.9 million, marking the fourth consecutive quarter of positive cash flow. The company's total assets reached $2.2 billion, nearly tripling from $759 million in the same period last year.

[Trump Media Reports $406 Million Net Loss in Q1, Unrealized Losses from Crypto and Equity Dominate]On May 9, Trump Media & Technology Group (TMTG) released its financial report for Q1 2026, showing a net loss of $405.9 million, with unrealized losses from digital assets and equity securities amounting to $368.7 million, making up the majority of the total loss. According to CoinGecko data, TMTG's crypto asset portfolio is currently valued at $821.9 million, with a cost basis of $1.24 billion, resulting in an overall unrealized loss of approximately $423 million. The asset portfolio includes 9,542 BTC (valued at $767 million, with an average price of $118,529) and 756 million CRO (valued at $54 million). Bitcoin fell about 22% in the quarter, marking the worst quarterly performance since 2018. On the operational side, Truth Social generated only $900,000 in revenue, but achieved an operating cash flow of $17.9 million, marking the fourth consecutive quarter of positive cash flow. The company's total assets reached $2.2 billion, nearly tripling from $759 million in the same period last year.

Trump Media Reports $406 Million Net Loss in Q1, Unrealized Losses from Crypto and Equity Dominate

On May 9, renowned KOL Hanbalongwang posted on X in response to the 'sat1 call event.' He stated that he participated solely as a trader and has no affiliation with the sat1 token project itself. The call for sat1 was only voiced within a group, and other groups amplified its influence through sharing, hoping to create liquidity. The reason for the call was to encourage sat1 to surpass sato for mutual profit, but he underestimated the level of PVP in the primary market, which failed to create a collective force. His current holdings are approximately at break-even, with a cost of 565 ETH, currently valued at 543 ETH. He stated the following regarding the handling of the call event: he will immediately sell all his sat1 holdings, and the ETH obtained will be used to compensate users who incurred losses from the call. He will no longer prove his profit or loss and will withdraw from the internet after the event is resolved. According to GMGN market data, as of the time of publication, sat1's market capitalization is reported at $1.2 million, having previously peaked at $8.4 million.

[KOL Hanbalongwang Responds to Sat1 Call Event: Will Sell Holdings to Compensate Users for Losses]On May 9, renowned KOL Hanbalongwang posted on X in response to the 'sat1 call event.' He stated that he participated solely as a trader and has no affiliation with the sat1 token project itself. The call for sat1 was only voiced within a group, and other groups amplified its influence through sharing, hoping to create liquidity. The reason for the call was to encourage sat1 to surpass sato for mutual profit, but he underestimated the level of PVP in the primary market, which failed to create a collective force. His current holdings are approximately at break-even, with a cost of 565 ETH, currently valued at 543 ETH. He stated the following regarding the handling of the call event: he will immediately sell all his sat1 holdings, and the ETH obtained will be used to compensate users who incurred losses from the call. He will no longer prove his profit or loss and will withdraw from the internet after the event is resolved. According to GMGN market data, as of the time of publication, sat1's market capitalization is reported at $1.2 million, having previously peaked at $8.4 million.

KOL Hanbalongwang Responds to Sat1 Call Event: Will Sell Holdings to Compensate Users for Losses

Join us for an exciting evening of blockchain networking with Astarter! Connect with 100+ investors and the community.

[Astarter Turkey Meetup — Istanbul Networking Night 🇹🇷]Join us for an exciting evening of blockchain networking with Astarter! Connect with 100+ investors and the community.

Astarter Turkey Meetup — Istanbul Networking Night 🇹🇷

On May 9, prediction market platform Kalshi announced a partnership with the sports venue Madison Square Garden. Kalshi has become its official prediction market partner, and the sixth-floor concourse of Madison Square Garden (primarily used for merchandise sales, dining, bars, and high-traffic areas) has been officially renamed the Kalshi Concourse.

[Kalshi Partners with Madison Square Garden as Official Prediction Market Partner]On May 9, prediction market platform Kalshi announced a partnership with the sports venue Madison Square Garden. Kalshi has become its official prediction market partner, and the sixth-floor concourse of Madison Square Garden (primarily used for merchandise sales, dining, bars, and high-traffic areas) has been officially renamed the Kalshi Concourse.

Kalshi Partners with Madison Square Garden as Official Prediction Market Partner

HTX data monitoring shows that ETH has fallen below 2300 USDT, currently reported at 2298.93 USDT, with a 24H decline of 2.87%.

【ETH falls below 2300 USDT】HTX data monitoring shows that ETH has fallen below 2300 USDT, currently reported at 2298.93 USDT, with a 24H decline of 2.87%.

ETH falls below 2300 USDT

$1,000 in Little Pepe ($LILPEPE) Could Mirror Early Ethereum Gains if Adoption Trends Continue

硅谷大厂的AI“含金量”开始分级

对话a16z Crypto：AI替你购物的时代，会变成什么样？

BVNK 创始人：稳定币发展的三个阶段

Bitcoin 2026大会洞察：挖矿衰落，AI崛起，一场结构性大迁徙

Sam Altman最新专访自白：其实我也不太懂AI内部在发生什么

对话Dragonfly合伙人：散户离场机构兜底，今年加密货币波动性难超美股

XRP Power Play: SBI Holdings Submits LOI To Acquire Bitbank In Major Asia Move

一周代币解锁：ENA解锁价值约1700万美元代币

This Week In Bitcoin: Top Developments That Could Signal A New Era

Article

Directly Competing with Claude Code: SpaceXAI's All-Platform Desktop Agent Grok Build Unexpectedly Revealed

ETH falls below 2300 USDT

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