Beam · 183

Meituan Chinchilla's core team evolves: pre-training and post-training leaders have left one after another

Comparative news, according to monitoring, Pei Peng, head of Meituan LongCat's Text-Based Model and Native Multimodality, has decided to leave his job and is currently no longer participating in team decisions. Su Hui, the head of LongCat post-training, has also left and joined another leading manufacturer. Pei Peng was previously responsible for the text base, native multimodality, and pre-training, and participated in the promotion of LongCat-2.0. Su Hui is responsible for aligning, fine-tuning, and reinforcement learning for the LongCat series models, and is also responsible for Agent Project Beam. The two each master the two key links before and after model training. Meituan denied Pei Peng's departure on July 27.

23d ago

India's Law Enforcement Authority raided multiple crypto fraud dens, targeting foreign investors

According to the anti-money laundering law (PMLA) from July 18 to 19, the Bengaluru branch of the Indian Law Enforcement Authority carried out raids on various places involved in the case to investigate and handle a virtual asset OTC transaction fraud case targeting overseas investors. Through social private channels, the criminal gang claims that they can supply tokens such as MultiversX, Kava, BEAM, GRASS, SUI, VANA, AGLD, etc. at discounted prices to induce foreign investors to invest. The total amount involved in the case was about 35 million US dollars. The case was initially reported and filed by a Dutch entity. The operation has already seized electronic devices, wallet documents, and cryptographic assets worth 8,700 USDT, and the law enforcement authorities are continuing to expand their clues and dig deeper into the complete criminal chain. This article is sponsored by GENG, Build Your Fortune on GENG (https://geng.one)

32d agoburnking

Crypto projects' flee 'their old names in bulk: the liquidity reset game behind brand upgrades

Author: Gu Yu, ChainCatcher Original title: Why do crypto projects always like to change their names? In the traditional business world, brand assets are the lifeblood of an enterprise. Frequent name changes are almost tantamount to actively destroying a moat. Nvidia won't change its name every few years, Apple won't give up on Apple because of some kind of business transformation, and Nike won't bring back the brand because of a sluggish market cycle. But in the cryptocurrency world, the rules are often the opposite. According to RootData statistics, more than 16% of encryption projects have changed their names, and many well-known first-line projects have also changed their names in large numbers. Just yesterday, the on-chain IP ecosystem Story Protocol announced that it will change its name to DATA, and IP tokens will migrate 1:1 to new DATA tokens. Within a few months, Xion changed its name to Verona, Matrixport changed its name to BIT, and TON's token symbol to GRAM. Earlier, a number of well-known projects such as Klaytn, EOS, Fantom, MakerDAO, Elrond, and Matic Network changed their names. More extreme projects have even changed their names more than once. For example, MAITRIX used names such as CENTRAL, X Network, and XLD Finance; BitSafe used the names dlcBTC and DLC.Link; Talex used the names Read2N and Metale Protocol; and KGen used the names IndigG and Kratos Gaming Network. The names have changed more and more, but most projects have not gained new life due to the new name; instead, they have gradually fallen silent. This brings up a question that is rarely seriously discussed in the crypto industry: Why do crypto projects always like to change their names? The answer is probably not complicated: because in the crypto industry, brands aren't the most important assets; attention, narrative, token prices, and liquidity are. 1. Crypto brand loyalty is too low. The reason traditional brands are afraid to change their names is because user loyalty comes from long-term consumer experiences. A user has bought an iPhone for many years, drank Starbucks for many years, and worn Nike for many years. His perception of the brand was not formed in a day, nor did it change easily due to a certain marketing campaign. But cryptographic projects have a completely different user structure. Most early users aren't consumers in the traditional sense, but investors, airdrop hunters, liquidity providers, node participants, and narrative traders. They use products not necessarily because they are easy to use, but because they may have air investment, may be profitable, and may have room for growth. This means that crypto brands are naturally less loyal to users. In the traditional industry, users ask “Is this brand worth trusting”; in the crypto industry, users are more often asked “can this coin rise?” As long as prices are sluggish for a long time, the narrative fails, and the ecology is silent, the old name will instead become a negative asset. A name that has experienced a crash, duvet cover, hacking, team controversy, or route failure can hardly inspire the market's imagination. It doesn't carry brand assets, but K-line scars and community grievances. This is the root reason why crypto projects dare to change their names frequently: in many cases, old names have no moats, only historical baggage. 2. Renaming is a marketing strategy. Not every name change should simply be viewed as a “vest change.” The name change of some projects is indeed because the original name cannot carry the new strategic scope. As hot market concepts change, if the name includes old concepts such as “Social” and “DAO,” or if the meaning of the name does not match, changing the name is an inevitable choice. For example, the decentralized social networking protocol OpenSocial changed its name to Eden after transforming AI, the decentralized electronic signature platform EthSign chose to remove “Eth” from its name after expanding its business, and the Ethereum sidechain Matic Network changed its name to Polygon (meaning polygon) after building multiple scaling solutions. When the project's business boundaries fundamentally change, the original brand may limit external perception. The name change is a necessary strategic calibration at this point. Of course, there are also quite a few projects that actively “grab hot spots”, and you can get more attention by naming popular concepts. In the last metaverse boom, Elrond changed its name to MultiversX and directly added “Multiverse” elements to the name, apparently hoping to join Yuanyu...

57d agoburnking#encryption #Exchange coins

Introducing AlphaGo Search, a new MCTS video generation framework with a longer video duration than Sora

Comparative news, according to monitoring, researchers from institutions such as the University of Waterloo and Brown University proposed a new Test Time Scaling (Test-Time Scaling) framework called Planning at Inference in a paper submitted in ICLR 2026. For the first time, AlphaGo's Monte Carlo Tree Search (MCTS) algorithm was applied across borders to long video generation. This framework models long video generation tasks as sequential decision problems. The system introduces MCTS in the inference stage and uses look-ahead rollouts (look-ahead rollouts) and backpropagation rewards to evaluate various video continuation clips, fundamentally solving the problems of semantic drift and error accumulation commonly faced in traditional chunked or single generation. To enable efficient exploration in a continuous video generation space, the research team specially designed the Multi-Tree MCTS (Multi-Tree Monte Carlo Tree Search) variant. Compared with the traditional method of using a single search tree under a fixed computing power budget, the multi-tree architecture can perform extensive searches in a continuous state space with more reasonable pruning and branching coefficients, significantly improving exploration efficiency. More importantly, Planning at Inference is extremely modular and is a completely plug-and-play inference optimization scheme. Developers can deploy this solution directly to existing video generation platforms without any retraining or fine-tuning of the underlying large model. In experiments based on Nvidia's open source video prediction model Cosmos-Predict2, Planning at Inference showed strong generation performance. In the long video generation evaluation, this solution successfully generated more than 20 seconds of high-quality coherent video. Test data showed that MCTS search generation quality was greatly improved compared to traditional baseline methods such as greedy search (Greedy Search), beam search (Beam Search), and best-of-N in terms of core indicators such as object persistence, temporal coherence, and text-to-video alignment. Compared to the current industry-leading large closed source model, the videos generated by this method are 18% longer than Sora and 47% longer than Kling, respectively, while maintaining the same level of image detail and visual fidelity as the two. Although the search mechanism brings excellent picture coherence, the introduction of multi-tree searches during the inference stage also brings high computational power costs. Researchers confess that the current Planning at Inference framework is significantly slower in terms of generation speed than traditional autoregressive direct generation, which limits the possibility of real-time deployment to a certain extent. However, as the efficiency of the underlying video generation platform evolves and computational hardware computing power continues to grow, the inference scaling route of calculating costs in exchange for image quality is expected to become a key technical path for long video generation to be practical in engineering after the basic capabilities of large models break through a specific threshold.

87d ago

Coinbase suspends 25 perpetual contract transactions and completes automated settlement

Comparative news: According to official sources, Coinbase has suspended 25 perpetual contract transactions as previously announced, and related open positions have been automatically settled at the final settlement price. It involves TRB, RARE, NEIRO, A, ME, XTZ, KMNO, RAY, STX, ENS, GMT, SNX, 1000FLOKI, 0G, ORDI, NIL, BIO, UMA, BEAM, INIT, SOMI, EGLD, CLANKER, SOPH, BIGTIME, etc. The final settlement price is calculated based on the average index price 60 minutes before the suspension of trading. Coinbase said the move aims to focus on products that continue to meet liquidity and market quality standards, and will accelerate the launch of new perpetual contracts by optimizing internal processes.

123d ago

Arweave AO launches network availability pledge test plan to launch gateway data service incentive mechanism

According to official news, AO, a scalable blockchain network based on Arweave, announced the launch of the Network Availability Staking Alpha (NASA) test program. This is a key step in the AO ecosystem, which aims to improve the usability and reliability of decentralized data networks through a staking mechanism. The program is currently in alpha, and users can participate in network availability verification and earn rewards by providing data services to Arweave gateways and staking AO tokens. In the first pilot phase, AO introduced an availability staking mechanism, requiring node operators to stake 25 AOs to participate in the network and compete for speed and stability in response to user requests. The system will distribute rewards from a monthly reward pool of 1000 AOs based on the node's performance in the data service. The mechanism relies on the next-generation HyperBeam architecture to enable gateways and routing services to achieve a higher degree of verifiability and trustlessness, while significantly reducing operating costs. The project party said that NASA aims to establish a stronger decentralized economic model for the entire permanent network infrastructure, paving the way for future expansion into fields such as computational scheduling, data indexing, and network services. Although the current rewards are small and still in the testing phase, the program is seen as an important starting point for the AO-core economy and will gradually expand to more network infrastructure services over time.

128d ago

Data: GMGN smart money 24h net inflow list, Triplet tops

According to GMGN data, the net inflow of smart money into the top 5 tokens in the past 24 hours is as follows: 1. Triplet (J8PS... ump): Net inflow of $7,000, up -23.4% in the past 24 hours, now at $0.0015. 2. AUTISM (8jiV... ump): Net inflow of $4,000, up -22.6% in the past 24 hours, now reported at $0.0031. 3.Punch (NV2R... ump): Net inflow of $3,000, up -11.2% over the past 24 hours, now reported at $0.0123. 4.Baby (BeAM... ump): Net inflow of $1,000, up 92.1% over the past 24 hours, now at $0.0004. 5. Swarms (74SB... ump): Net inflow of $1,000, up 5.6% in the past 24 hours, now at $0.0075.

171d ago

Data: GMGN smart money 24h net inflow list, PENGUIN tops

According to GMGN data, the net inflow of smart money into the top 5 tokens in the past 24 hours is as follows: 1. PENGUIN (8Jx8... ump): Net inflow of $5,000, up -29% in the past 24 hours, now at $0.0277. 2.BEAM (CV38... rmJ): Net inflow of $3,000, up 139.1% over the past 24 hours, now at $0.0044. 3.x (4P63... bA): Net inflow of $2,000, up -16.7% in the past 24 hours, now at $0.0001. 4. KITTY (dFFp... ump): Net inflow of $1,000, up 169.9% over the past 24 hours, now at $0.0006. 5.SOLO (Solo... eta): Net inflow of $695, up -2.3% in the past 24 hours, now reported at $0.5117.

199d ago

Data: The crypto market generally declined, the GameFi sector led the decline of nearly 5%, and BTC fell below $88,000

Comparative news, according to SosoValue data, the crypto market sector generally declined, with GameFi leading a 24-hour decline of 4.90%. Within the sector, The Sandbox (SAND) fell 8.85%, Axie Infinity (AXS) fell 18.23%, but Beam (BEAM) bucked the trend and rose 19.02%. Meanwhile, Bitcoin (BTC) fell 1.84% to below $88,000; Ethereum (ETH) fell 2.34% below $2,900. In terms of other sectors, the CeFi sector fell 1.55% in 24 hours, with Aster (ASTER) falling 6.86%; PayFi falling 2.03% and Monero (XMR) falling 10.25%; Meme sector falling 2.10% and PIPPIN (PIPPIN) falling 18.01%; Layer1 sector falling 2.25%, with TRON (TRX) being relatively strong, up 0.34%; DeFi sector fell 3.10% River (RIVER) rose again by 30.71% intraday; the Layer 2 sector fell 4.63%, but Movement (MOVE) rose 2.38%. The crypto sector index, which reflects the sector's historical market, shows that the SSIGameFi, SSIlayer2, and SSIAI indices fell 6.05%, 4.93%, and 4.39%, respectively.

208d ago

Data: The GMGN KOL list shows that VVM has received much attention and obtained multiple KOLs net inflows

Comparative news, according to GMGN data, the net KOL inflow to the top 5 tokens in the past 24 hours is as follows: 1. VVM (hWBn... AGS): Net inflow of $5,000, up 4201.3% in the past 24 hours, now at $0.0003. 2.Baby (BeAM... ump): Net inflow of $4,000, up 55.5% in the past 24 hours, now at $0.0005. 3. MICHAEL (7WYN... AGS): Net inflow of $3,000, up 3986.3% in the past 24 hours, now at $0.0002. 4. GRANNY (ewbp... ump): Net inflow of $3,000, up 219.2% over the past 24 hours, now at $0.0002. 5.CHEETAH (5rhB... AGS): Net inflow of $2,000, up 2517.2% over the past 24 hours, now at $0.0002.

225d ago