Done!This launches our 100% locally running mini-ChatGPT that is powered by DeepSeek-R1.Check this demo 👇 https://t.co/CQ6mOAj8Rt ...

Let's build a mini-ChatGPT app powered by DeepSeek-R1 (100% local): ...

不圆 发自 凹非寺 量子位 | 公众号 QbitAI DeepSeek推理要详细还是要迅速，现在可以自己选了？ 来自特拉维夫大学的研究团队开发出了一种新方法，可以 监控和控制LLM中的思考路径长度 。 超频能够减少不必要的推理步骤，使模型更快地得出结论，同时避免因过度推理导致的性能下降。 该模型已在gitHub上开源。 给LLM的推理任务装上进度条，还能控制推理的深度、调整推理速度。 加速后的模型和原模型相比， 使用的token数减少了近6倍，且都得出了正确答案 。 LLMs在显示结构化推理时，会隐式跟踪其在思考阶段的相对位置，并通过隐藏状态编码这一信息。 而论文提出了一种"思维进度向量"（Thinking Progress Vector, TPV ），可用于实时预测模型在推理阶段的相对位置，并通过可视化进度条 展示模型的推理动态。 通过干预TPV，可以加速或减速模型的推理过程，实现"超频"（overclocking）和"降频"（downclocking）。 方法：实时监控并控制推理深度 在有效推理学习过程中，模型必须 隐式地学习跟踪其思考阶段进度 ，并保持对例如距离最终答案有多近的估计。 由于进度跟踪依赖于 ...

Core Viewpoint - The article discusses the evolution of DeepSeek in the context of Mixture-of-Experts (MoE) models, highlighting innovations and improvements from DeepSeekMoE (V1) to DeepSeek V3, while maintaining a focus on the MoE technology route [1]. Summary by Sections 1. Development History of MoE - MoE was first introduced in 1991 with the paper "Adaptive Mixtures of Local Experts," and its framework has remained consistent over the years [2]. - Google has been a key player in the development of MoE, particularly with the release of "GShard" in 2020, which scaled models to 600 billion parameters [5]. 2. DeepSeek's Work 2.1. DeepSeek-MoE (V1) - DeepSeek V1 was released in January 2024, addressing two main issues: knowledge mixing and redundancy among experts [15]. - The architecture introduced fine-grained expert segmentation and shared expert isolation to enhance specialization and reduce redundancy [16]. 2.2. DeepSeek V2 MoE Upgrade - V2 introduced a device-limited routing mechanism to control communication costs by ensuring that activated experts are distributed across a limited number of devices [28]. - A communication balance loss was added to address potential congestion issues at the receiving end of the communication [29]. 2.3. DeepSeek V3 MoE Upgrade - V3 maintained the fine-grained expert and shared expert designs while upgrading the gating network from Softmax to Sigmoid to improve scoring differentiation among experts [36][38]. - The auxiliary loss for load balancing was eliminated to reduce its negative impact on the main model, replaced by a dynamic bias for load balancing [40]. - A sequence-wise auxiliary loss was introduced to balance token distribution among experts at the sequence level [42]. 3. Summary of DeepSeek's Innovations - The evolution of DeepSeek MoE has focused on balancing general knowledge and specialized knowledge through shared and fine-grained experts, while also addressing load balancing through various auxiliary losses [44].

Core Viewpoint - The article discusses the increasing prevalence of misinformation generated by AI, highlighting the challenges posed by AI hallucinations and the ease of feeding false information into AI systems [3][10][21]. Group 1: AI Misinformation - AI hallucination issues lead to the generation of fabricated facts that cater to user preferences, which can be exploited to create bizarre rumors [3][10]. - Recent examples of widely circulated AI-generated rumors include absurd claims about officials and illegal activities, indicating a trend towards sensationalism over truth [5][6][7][8]. Group 2: Impact of Social Media - The combination of AI's inherent hallucination problems and the rapid dissemination of information through social media creates a concerning information environment [13][14]. - The article suggests that the current state of information is deteriorating, likening it to a "cesspool" [15]. Group 3: Recommendations for Improvement - AI companies need to enhance their technology to address hallucination issues, as some foreign models exhibit less severe problems [17]. - Regulatory bodies should improve their efforts to combat the spread of false information, although the balance between regulation and innovation remains delicate [18]. - Individuals are encouraged to be cautious with real-time information while relying on established knowledge sources [20].

Group 1 - Apple is considering a significant shift in its AI strategy, potentially moving away from developing its own large language models to utilizing OpenAI's ChatGPT or Anthropic's Claude models for Siri [5] - Nvidia's market capitalization approached $4 trillion, briefly surpassing Apple's previous record, with a stock price increase of 17.92% since June [3] - Meta has established a new department called "Meta Superintelligence Lab" (MSL), led by former Scale AI CEO Alexandr Wang, and has recruited several key personnel from OpenAI and Anthropic [4] Group 2 - Huawei's Pangu team denied allegations of plagiarism regarding their open-source model, stating that their Pangu Pro MoE model was developed independently on their Ascend hardware platform [2] - Both Baidu and Huawei announced their latest open-source models on June 30, with Baidu releasing ten models from its Wenxin series and Huawei open-sourcing models with parameters up to 720 billion [7] - xAI, founded by Elon Musk, secured $10 billion in new funding, which includes $5 billion in debt and $5 billion in equity, to support its AI development initiatives [8] Group 3 - OpenAI's CEO criticized Meta's recruitment practices, expressing concerns about potential cultural issues within companies due to talent poaching [9] - Ilya Sutskever announced his appointment as CEO of Safe Superintelligence (SSI) after the departure of co-founder Daniel Gross, who joined Meta's Superintelligence Lab [10][11] - The price of DDR4 memory modules has nearly doubled in the past month due to supply constraints and increased demand for AI-related applications [13]

Group 1 - Pangu Pro MoE model is developed based on Ascend hardware platform and not incrementally trained from other vendors' models, adhering to open-source licensing requirements [3] - DeepSeek did not issue an apology regarding the false association with Wang Yibo, and the alleged apology statement was AI-generated [5] - Zhaowenqi, an actress from "iPartment," claims to have been deceived by a partner in her business venture, leading to financial losses [6] Group 2 - Romoss announced a six-month suspension of operations starting July 7, 2025, due to market changes, with initial salary payments for employees [6] - Meituan's daily orders exceeded 120 million, with over 100 million coming from the food delivery sector, marking a significant increase during the summer consumption peak [6] - WeChat's HarmonyOS version development is slower than expected, with the team focusing on foundational work before adapting to the new system [6] Group 3 - Sales of 3C certified power banks surged following a recall incident, with many merchants facing stock shortages [8] - Shanghai Lego Park, the largest in the world, opened on July 5, covering 318,000 square meters and featuring over 75 interactive attractions [8] - Chinese electric vehicles dominated the Israeli market in the first half of 2025, with 21,252 units sold, accounting for 81.2% of total electric vehicle sales [10] Group 4 - Xiaomi's YU7 model began deliveries across 58 cities on July 5 [10] - Chuangbu Technology completed a 32 million RMB Series A financing round to expand its digital payment platform [10] - Jingxiang Technology secured several million in Pre-A financing to enhance AI technology and brand development in the esports sector [10] Group 5 - BYD's first hybrid travel car, Seal 06DM-i, was launched with a price range of 109,800 to 129,800 RMB and features intelligent driving capabilities [11] - ZhiYuan Robotics showcased its X2-N model, highlighting its all-terrain mobility and innovative design inspired by Chinese mythology [13] - Porsche's first electric Cayenne features a high-tech interior with four screens and minimal physical buttons, expected to launch in 2026 [15][17] Group 6 - Lynk & Co's 10 EM-P debuted as a luxury sports sedan with standard four-wheel drive and laser radar, targeting the mainstream market [17] - Tesla's sales in the UK increased by 14% in June, driven by rising demand for electric vehicles [19]

Core Insights - The article emphasizes the imminent arrival of the humanoid robot era, as highlighted by industry leaders like Nvidia and Tesla during CES 2025, marking a significant shift towards embodied intelligence in 2025 [3] - Major investment banks, Morgan Stanley and Goldman Sachs, have released reports affirming the vast potential of the humanoid robot market, projected to be a trillion-dollar industry, while also detailing the current technological barriers and commercialization challenges [3][4] Investment Trends - The humanoid robot financing landscape is experiencing a bifurcation, with capital increasingly favoring leading startups, as evidenced by significant funding rounds such as The Bot Company's $150 million and Neura Robotics' €120 million [4][5] - In the first half of 2025, global financing in the embodied intelligence sector approached 20 billion yuan, indicating robust investment activity despite a two-tiered market dynamic [4] Company Valuations and Market Dynamics - Leading companies like Zhiyuan Robotics and Yushu Technology have surpassed valuations of 10 billion yuan, with Zhiyuan Robotics valued at approximately 15 billion yuan and Yushu Technology at around 12 billion yuan [7] - The majority of startups in the humanoid robot sector are valued below 3 billion yuan, indicating a highly competitive environment with uncertain prospects for many emerging companies [7] Commercialization Challenges - Despite Yushu Technology's reported annual revenue exceeding 1 billion yuan, the actual commercialization scenarios for humanoid robots remain questionable, with concerns about the sustainability of their business models [9][10] - The article notes that many humanoid robot companies are focusing on educational and research applications rather than industrial use, which may limit their long-term market viability [10] Technological Developments - The humanoid robot sector is entering a phase where companies are expected to develop their own "brains," with a shift towards self-research in foundational models becoming evident [12] - The introduction of DeepSeek's open-source reasoning model presents new opportunities for startups to break free from reliance on tech giants' closed-source models, potentially reshaping the competitive landscape [15] Supply Chain Insights - Upstream core component manufacturers in the humanoid robot supply chain are already reaping benefits, with stock prices of related companies experiencing significant increases [16] - The article highlights the importance of advanced components like dexterous hands and tactile sensors, which are critical for the commercialization of humanoid robots [17]

Group 1 - DeepSeek did not issue an apology to Wang Yibo regarding the AI model's inappropriate association with a corruption case, despite widespread rumors [1] - Lei Jun acknowledged that the price of the Xiaomi YU7 car-mounted tissue box is relatively high at 169 yuan, but emphasized the product's design considerations for extreme temperature conditions [1] - Gree Electric's market director clarified that the company values talent based on innovation and responsibility rather than educational background, countering previous statements by CEO Dong Mingzhu about not hiring overseas returnees [2] Group 2 - A Chinese model who was reported missing after being lured to Myanmar under the pretense of a modeling job has been rescued, highlighting the dangers of overseas job scams [2] - Taobao customer service reported that the seller Romashi is currently unable to process refunds due to insufficient account balance, following a recall of certain power bank models due to safety concerns [3][4] - The CEO of Yunhaiyao admitted legal responsibility for a food poisoning incident involving ByteDance employees in Singapore, where 60 employees fell ill after a company lunch [9] Group 3 - Nintendo's president defended the pricing of the Switch 2, which has increased by 100-200 USD compared to previous models, stating that the price reflects the gaming experience offered [8] - The Ministry of Industry and Information Technology in China announced a pilot program for number protection services, allowing users to choose whether to authorize their phone numbers for use on internet platforms [8] - Counterpoint Research reported that iPhone sales in China grew by 8% year-on-year in Q2, marking Apple's first sales increase in two years, with Huawei and Vivo leading the market [12]

Core Viewpoint - The venture capital market is showing signs of recovery, supported by objective data rather than subjective feelings, as key metrics from the first half of the year indicate a positive trend [1][2]. Group 1: Market Recovery Indicators - The scale of institutional LP (limited partner) investments surged by 50% year-on-year, while the decline in financing scale has significantly narrowed, and the number of IPO exit projects increased by over 20% [2]. - Multiple core indicators have rebounded collectively, marking the venture capital market's transition into a recovery cycle [2]. - A series of policy measures, including the new "National Nine Articles" and "Seventeen Articles on Venture Capital," are aimed at enhancing the support for technological innovation and facilitating the entire fundraising, investment, management, and exit chain [2]. Group 2: Investment and Funding Dynamics - Investment activity has notably increased, with AI and humanoid robotics companies like DeepSeek and Yushutech emerging as new hotspots for hard technology investments, leading to intensified competition for quality projects [2]. - Long-term capital is entering the market, exemplified by the National Big Fund's investment of nearly 200 billion yuan to establish three equity funds, alongside accelerated fundraising processes for state-owned and specialized funds [2]. - The exit landscape is structurally improving, with heightened activity in the Hong Kong IPO market and an increase in the quality and quantity of merger and acquisition cases [2]. Group 3: Challenges and Future Outlook - The core logic behind the recovery in fundraising and investment is the restoration of secondary market valuations and improved exit expectations [3]. - There is a growing consensus on the need for diversified exit mechanisms, with venture capital institutions focusing on enhancing DPI (Distributions to Paid-In) as a primary goal [3]. - However, the overall market recovery still faces challenges, such as the need to further activate market-based funding sentiment and expand the scale of long-term capital entering the market [3].