World Liberty Financial (WLFI), the Trump-linked DeFi project, is scrambling to stop a market collapse after its token lost over 50% of its value in September. On Friday, the project unveiled a full buyback-and-burn program, directing all treasury liquidity fees to absorb selling pressure. According to a governance post on X, the community approved the plan overwhelmingly, with WLFI pledging full transparency for every burn. The urgency of the move reflects WLFI’s steep losses in recent weeks. WLFI is tr ...

Decentralised exchanges offering perpetual trading for cryptocurrencies, such as Hyperliquid and Aster, are taking the industry by storm. The sector has become the centrepiece of crypto’s latest mania, accumulating $1.8 trillion in trading volume this quarter alone, according to DefiLlama data. That’s more than the niche’s entire trading volume recorded in 2024. Perp DEXs are decentralised exchanges for trading perpetual futures. These platforms allow traders to speculate on crypto price moves without ...

KEY TAKEAWAYSCrypto launchpads (e.g., Binance Launchpad, DAO Maker, Polkastarter, Seedify, Best Wallet) provide early access to ICOs, IEOs, and IDOs with project vetting and token distribution to catch gems before they launch.Aggregator sites & calendars like CoinMarketCap, CoinGecko, and CryptoRank track upcoming token launches, ICOs, tokenomics, and exchange listings, giving investors a catalog of fresh projects to observe and study.Blockchain developer platforms such as GitHub and tools like Token Termin ...

Ronin, the blockchain behind the hit game Axie Infinity, has announced a new plan to strengthen its token economy through smart moves that have surged the RON price. Is it a good time to buy RON? Let’s dive in. The Sky Mavis-built network said on Sept. 21 that it will begin buying RON from the open market starting Sept. 29. Over about a month, the team will convert its Treasury’s 890 ETH and 650,000 USDC, roughly $4.5-$5M, into RON. At current prices, that equals about 1.3% of circulating supply. All pu ...

A federal jury’s conviction of SafeMoon CEO Braden John Karony on fraud and money-laundering charges has heightened U.S. scrutiny of token promoters, as the FBI seeks investors defrauded in the collapsed DeFi project. Last week, the FBI opened a victim questionnaire, asking SafeMoon investors who lost money to submit information that could support restitution and help identify the full scope of the fraud. Karony, 29, was found guilty in May after a two-week trial in Brooklyn, where prosecutors showed he and ...

纪要涉及的公司和行业 - **公司**：DeepSeek、OpenAI、Anthropic、Google、Microsoft、Amazon、Parasail、Friendli、Lambda、Nebius、Cursor、Oracle、Huawei、AMD、NVDA等 [4][22][24][29][30] - **行业**：人工智能（AI）行业，特别是大语言模型（LLM）领域 [4] 纪要提到的核心观点和论据 DeepSeek相关 - **核心观点**：DeepSeek R1发布后虽有影响，但自身服务市场份额下降，原因在于其在tokenomics上的权衡导致用户体验不佳 [8][13][21] - **论据** - **价格与延迟**：DeepSeek虽产品价格便宜，但用户需等待数秒模式才响应首个token，相比其他提供商延迟高，如Parasail、Friendli等可提供低延迟服务，微软Azure价格高但延迟低25s，且多数R1 0528实例现以低于5秒延迟托管 [22] - **上下文窗口**：DeepSeek运行K上下文窗口，是主要模型提供商中最小的之一，限制了如编码等需要大上下文窗口的用例，而相同价格下其他提供商如Lambda和Nebius可提供2.5倍上下文大小 [24] - **市场份额变化**：发布后消费者应用流量激增，但未跟上其他AI应用增长，自身网络应用和API服务市场份额下降，而第三方托管实例使用量近20倍增长，自身总token份额每月下降 [8][9][10][13] - **持续发展**：R1发布后持续扩展强化学习（RL），在许多领域尤其是编码方面有所改进，推理模型不断快速改进且更有效、更实惠 [5][7] Anthropic相关 - **核心观点**：Anthropic在编码应用方面成功，但受计算资源限制，不过用户体验优于DeepSeek [29][32][41] - **论据** - **编码应用成功**：Anthropic专注代码产品开发，其Claude Code使用量激增，超过OpenAI的Codex，Google也推出类似工具Gemini CL [29][30] - **计算资源压力**：Claude的输出速度在API上下降30%至略高于5 tokens每秒，因需处理大量请求且编码使用倾向大token数对话，相比之下OpenAI和Google的模型速度更快 [33] - **用户体验优势**：速度虽低但快于DeepSeek的2.5 tokens每秒，且回答问题所需token数远少于其他模型，端到端响应时间更低，Claude在领先推理模型中总输出token数最少 [41][42] 市场动态相关 - **核心观点**：AI市场竞争激烈，各公司在计算资源、价格、性能等方面竞争，开源模型有发展潜力，速度可通过其他因素补偿 [44][47][40] - **论据** - **价格竞争**：DeepSeek低价冲击市场后，OpenAI旗舰模型价格下降80%，其6月对GPT - 4 API定价大幅削减，缩小与R1的价格/性能差距 [4][53] - **计算资源投入**：Amazon大规模建设AI集群，投资数十亿美元用于Trainium AI集群，Anthropic将获得超50万个Trainium芯片用于推理和训练，Anthropic还从Google GCP租用大量计算资源，Google云也向其他AI公司扩展服务 [37][38][39] - **开源模型潜力**：廉价计算资源的可用性和软硬件的快速创新将推动开源模型发展，如DeepSeek R1在编码性能上不断改进，对采用有积极影响 [47] 其他重要但是可能被忽略的内容 - DeepSeek将研发团队从杭州迁至北京，运营人员翻倍以应对媒体请求，虽有R2延迟传言，但并非因出口管制导致训练延迟，且其仍保持快速招聘速度，还为华为盘古模型提供技术支持 [54][55][56] - 推理云兴起，越来越多公司效仿Anthropic以服务形式销售token，而非像ChatGPT那样以月度订阅形式捆绑销售 [44]