"有时候，真的被哈基米萌得哈特（heart）软软！""调理好的哈基米也太香了！" 如果你最近在小红书、微博上看到这类分享，别误会，大家讨论的不是猫，而是谷歌的AI大模型—— Gemini。 给AI起外号不稀奇，DeepSeek被叫"D老师"，Claude被叫"克劳德"，都还算正常。但"哈基米"这个称呼，透 着一股强烈的偏爱和宠溺，显得格外不同。 在AI圈内，这是一个很有趣的现象：一边，是谷歌在开发者大会上，"原生多模态"、"架构"、"延迟优 化"，努力将Gemini塑造成一个强大、可靠的生产力工具；另一边，是用户在小红书、SillyTavern和贴吧 里，把它当成"哈基米"、"猫猫"和需要"调教"的"逆子"。大家关心的不是模型参数，而是"怎么写prompt才 能让它不哈气"、"今天的哈基米心情不好，一句话怼我三次"。 技术严肃性与社区娱乐性之间正存在着巨大反差。 1 "哈基米"是怎么叫起来的？ 但语言有种奇妙的魔力，当"Gemini"被念成"哈基米"时，这个源于网络、充满宠溺感的猫咪梗，便不由分 说地为这个AI披上了一层情感滤镜。很快，"芥末泥"、"小gem"之类的爱称也层出不穷。 而Gemini自身的模型特 ...

Great to have you here. Hi everyone. I'm Chad Nelson. I'm a creative director here at OpenAI.Stop. Stop. Stop.We'll keep it casual, but no, this is going to be a fun talk. Uh so in my career I've made a lot of content, films, video games, television shows and in all my experiences there's been one common kind of through line which is how am I going to make it. How am I going to make this stuff ultimately like what are my creative workflows that I need. What are the tools I'm going to use. And so what happen ...

Earlier this year, a tweet went viral with a new term, vibe coding. Now, it got a lot of polarizing reactions, but I think one thing we can all agree on is it really changed the way we saw how AI was going to transform software engineering and how products are built. But we're not here today to talk about buzzwords or hype or what might be possible.Today, we are here to show you startups that our team has been working closely with and we have seen firsthand how their AI coding tools have helped tiny teams 1 ...

Hey everyone, how's it going. How we feeling. Do we like the keynote.Good. Uh my name is Scotty. Uh super excited to be here today.Uh you just heard on the main stage about new features that are going to make it easier to deploy agent-driven workflows. In the next 20 minutes, we're going to talk about how we put that technology to work inside open AAI. I'm joining from the go to market innovation team.We are a team of developers that are responsible for deploying our technology inside our goto market organi ...

Maral Libra is by far the largest e-commerce and fintech platform in Latin America. [Music] We're helping sellers deal with complicated situations that are even complicated for human beings. The ability to use the right tool with the right parameters is absolutely key to scale.So for example, building an agent, it's a complex endeavor. You have to have conversations and reason and use tools. But the great thing about GPT5, it combines the ability to have conversation and reasoning on the same model.Before t ...

Core Viewpoint - OpenAI CEO Sam Altman predicts that general artificial intelligence (AGI) could be achieved by 2030, with capabilities far exceeding human intelligence, referred to as "super AI" [1][4]. Group 1: Development of AGI - Altman expresses confidence in the rapid advancement of AI technology, stating that the current GPT-5 model already surpasses many human capabilities [1]. - He emphasizes that if humanity has not developed extraordinary models by 2030, he would find it "hard to understand" [1]. Group 2: Impact on Employment - Altman addresses concerns about AI taking over jobs, suggesting a rational perspective that the job market is already dynamic and evolving [4]. - He estimates that AI could potentially take over 30% to 40% of jobs in the economy, indicating a significant shift in employment structures [4]. - He advises teaching children the fundamental ability to learn and adapt to changes, highlighting the importance of lifelong learning and human uniqueness in understanding needs and creating value [4]. Group 3: Human-AI Relationship - On the topic of whether AI could pose a threat to humanity, Altman maintains a balanced view, suggesting that AGI could treat humans with a "motherly" approach [4]. - He acknowledges the potential for AGI to produce unforeseen side effects and consequences, urging caution regarding the risks associated with AI technology [4]. Group 4: Personal Insights on AI Usage - Altman shares his personal approach to AI, indicating that he does not use ChatGPT for emotional issues, reflecting an understanding of AI's limitations in handling complex human emotions [5].

Core Viewpoint - Grok-4-fast has demonstrated exceptional performance in cost reduction and efficiency, surpassing even GPT-5, which is associated with routing capabilities [1][38]. Group 1: Performance and Efficiency - Grok-4-fast's impressive reasoning efficiency is attributed to advanced scaling of computational power [2]. - The underlying technology of Grok is linked to NVIDIA's algorithmic advancements, particularly a new model called Jet-Nemotron [3][4]. - Jet-Nemotron-2B has shown performance comparable to leading open-source models while achieving a speed increase of approximately 53 times [7]. Group 2: Technological Innovations - The key innovation behind Grok-4-fast is a new framework called PostNAS, which significantly reduces training costs and allows for more comprehensive exploration of model structures [10][11]. - PostNAS employs a hybrid structure model that retains essential attention layers while eliminating redundant ones to enhance efficiency [13][14]. - The framework includes four core components: full attention layer placement, optimal linear attention module selection, design of superior linear attention modules, and hardware-aware architecture search [12]. Group 3: Attention Mechanisms - The NVIDIA team evaluated six advanced linear attention modules, with Gated DeltaNet achieving the highest accuracy due to its data-dependent gating mechanism and delta rule [18][19]. - JetBlock, a more advanced linear attention module, utilizes dynamic convolution to adaptively generate convolution kernels based on input features, outperforming Gated DeltaNet in accuracy for mathematical reasoning and retrieval tasks [21][24]. Group 4: Hardware Optimization - NVIDIA's hardware-aware architecture search focuses on optimizing key parameters rather than solely relying on parameter size, which does not accurately reflect real hardware efficiency [27][28]. - The team found that the size of the key-value (KV) cache is crucial for throughput in long-context and long-text generation, leading to a targeted optimization approach [30][31]. Group 5: Industry Impact - PostNAS is expected to influence the AI industry by providing a low-cost, high-efficiency architecture exploration method applicable to any pre-trained transformer [34]. - The Jet-Nemotron model is open-source, allowing various manufacturers to integrate it without retraining, significantly reducing costs while maintaining accuracy [36][42]. - The potential application of Jet-Nemotron across major AI companies like OpenAI and Google could lead to widespread improvements in model performance and cost efficiency [43].

Core Insights - The article discusses the significant advancements in AI technology, particularly with the release of GPT-5 and DeepSeek-V3.1, highlighting the shift in market sentiment towards AI's transformative potential across various industries [1][2] - Keep has reported a substantial increase in revenue and profitability, achieving a revenue of 822 million yuan and a net profit of 10.35 million yuan for the first half of 2025, marking a significant improvement from previous periods [1][6] - The focus of AI applications is shifting from mere technological competition to empowering specific use cases, with companies like Keep leveraging AI to enhance user experience and engagement [3][4] Company Performance - Keep's revenue for the first half of 2025 reached 822 million yuan, with a gross profit of 429 million yuan and a gross margin increase from 46.0% to 52.2% year-on-year [1][6] - The adjusted net profit margin for Keep improved to 1.3%, a significant turnaround from a net loss margin of 15.5% in the same period last year [6] - The average monthly active users and monthly subscription members for Keep remained stable at 22.49 million and 2.8 million, respectively, indicating consistent user engagement [6] AI Integration and Market Trends - The article emphasizes that successful AI applications are not solely based on large models but also on understanding user needs and having relevant data, which enhances business logic and user engagement [3][5] - Keep launched Kinetic.ai, a specialized AI model for the sports and health sector, trained on data from 400 million users, allowing for personalized fitness plans [3][4] - The integration of AI has led to a more interactive user experience, with features like AI coaching that provide tailored fitness guidance and support [4][7] Industry Dynamics - The AI application market is approaching a tipping point, with companies like OpenAI recognizing the need to expand beyond conversational AI to unlock greater commercial value [5][8] - The shift towards vertical AI applications is seen as a way to drive business growth, as companies focus on core competencies and cut low-margin operations [8][9] - Keep's strategy involves a deep integration of AI into its services, aiming to provide personalized, intelligent, and scientific fitness experiences, thereby reshaping its competitive advantage [9][10]

Core Viewpoint - The U.S. software sector has been underperforming due to macroeconomic turbulence and slow AI integration, but the narrative of "AI destroying software" lacks fundamental support. The company predicts a rebound in the software sector similar to the second half of 2024, driven by interest rate cuts and the realization of AI revenue [1][9]. Group 1: Reasons for Underperformance - The weak performance of the U.S. software sector is attributed to multiple factors, including high volatility suppressing traditional software spending and concerns over economic stagnation [2][3]. - Traditional software companies are experiencing delays in AI product deployment, while startup AI firms are rapidly increasing revenue from a low base [3]. - The narrative of "AI consuming software" has intensified market fears regarding traditional software companies [3]. Group 2: AI Algorithm Progress - The average iteration speed of overseas large models has improved from six months to approximately four months since 2025, indicating rapid advancements in AI capabilities [4]. - Despite improvements, large models still have significant room for growth, particularly in complex tasks like system design and new feature development [4][5]. Group 3: Impact of AI on Software Sector - The core barriers in application software remain high-quality data and industry knowledge, leading to increased market concentration in complex business logic areas [6]. - The foundational software sector is expected to see growth driven by the need for data unification and governance as AI applications become more prevalent [6]. - The information security sector is experiencing increased demand for AI-driven protection solutions [6]. Group 4: Future Outlook - The current macroeconomic environment and performance cycles are similar to those expected in the second half of 2024, with anticipated interest rate cuts and clearer tariff frameworks [7]. - The deployment of AI products is expected to begin reflecting in revenues by Q3/Q4, with October's Q3 earnings reports serving as a key observation point [7].

Elon Musk and the XAI team are building a competitor to Microsoft Windows called MacroHard. And the premise is it will be an end-to-end neural network operating system. That means everything that you're familiar with with your operating system from Mac to Linux to Windows can be generated by artificial intelligence.This is a massive departure from what we know as a traditional operating system today. This is not the first time I've talked about this. I really do see a future in which everything video games, ...