熹妃，回宫！ 在全球用户的强烈呼声下，OpenAI 不得不让旧模型悉数回归，真是一出大戏啊。 Sam Altman @ @ @sama · 1h Updates to ChatGPT: You can now choose between "Auto", "Fast", and "Thinking" for GPT-5. Most users will want Auto, but the additional control will be useful for some people. 现在 GPT-5 Thinking 的速率限制为每周 3.000 条消息，超过后会在 GPT-5 Thinking mini 上提供额外容量。GPT-5 Thinking 的上下文限制为 196k 令牌。 我们可能会根据使用情况随时间调整速率限制。 4o 现在默认对所有付费用户在模型选择器中可用。如果我们将来要弃用它,会 提前充分通知。付费用户现在在 ChatGPT 网页设置中还有一个"显示更多模 型"开关,打开后会增加像 o3、4.1 和 GPT-5 Thinking mini 这样的模型。4.5 仅对 Pro 用户开放 ...

Core Insights - The article discusses the advancements in AI, particularly focusing on DeepMind's Genie 3 and its capabilities in creating a "world model" that understands physical laws [4][5][10] - The conversation highlights the rapid development pace at DeepMind, with new releases almost daily, indicating a significant momentum in AI research and applications [9][18][19] - The need for improved evaluation benchmarks for AI models is emphasized, as current models show inconsistent performance across different tasks [11][45][46] Group 1: Genie 3 and World Models - Genie 3 is designed to generate virtual worlds that operate in a realistic manner, aiming to create a comprehensive understanding of the physical world [4][5][33] - The model's ability to generate and interact with its own environments allows for innovative training methods, where one AI operates within another AI's generated world [38][39] - The development of Genie 3 is seen as a step towards achieving AGI, as it requires a deep understanding of physical interactions and behaviors [33][34] Group 2: DeepMind's Development Pace - DeepMind is experiencing a rapid release cycle, with significant advancements in AI technologies such as DeepThink and Gemini [15][19] - The excitement surrounding these developments is palpable, with internal teams struggling to keep up with the pace of innovation [18][19] - The focus on creating models that can think, plan, and reason is crucial for advancing towards AGI [10][25] Group 3: Evaluation and Benchmarking - There is a pressing need for new and more challenging evaluation benchmarks to accurately assess AI capabilities, particularly in understanding physical and intuitive reasoning [45][46] - The introduction of the Kaggle Game Arena aims to provide a platform for testing AI models in various games, which could lead to significant improvements in their performance [41][50] - The article suggests that traditional evaluation methods are becoming saturated, and innovative approaches are necessary to measure AI's cognitive abilities effectively [45][56]

Core Viewpoint - The AI application market is entering a new phase of growth, driven by advancements in models like GPT-5 and increasing demand for computing power, particularly in high-trust sectors such as healthcare, education, and finance [4][5]. Group 1: Market Performance - The Shanghai Stock Exchange Sci-Tech Innovation Board Artificial Intelligence Index (950180) rose by 0.43%, with notable increases in constituent stocks such as Jingchen Co., Ltd. (688099) up 7.62% and Youkede (688158) up 2.21% [3]. - The Sci-Tech AI ETF (588790) has seen a 2.82% increase over the past week, with a current price of 0.66 yuan [3]. - The latest scale of the Sci-Tech AI ETF reached 70.34 billion yuan, marking a new high since its inception [6]. Group 2: Investment Trends - The second phase of domestic AI application investment is underway, with a focus on hardware and multi-modal applications [4]. - The index is projected to achieve a net profit of 12.8 billion yuan in 2025, reflecting a year-on-year growth of 96.34% [4]. - The top ten weighted stocks in the index accounted for 67.36% of the total, indicating a concentrated investment in leading AI companies [10]. Group 3: Fund Performance - The Sci-Tech AI ETF experienced a significant increase in shares, with a growth of 3.63 million shares over the past week [7]. - The fund has shown a net inflow of 348 million yuan over the last five trading days, averaging a daily net inflow of 69.67 million yuan [7]. - The fund's performance metrics include a 5.60% increase in net value over the past six months, ranking first among comparable funds [8][9].

AI & Technology - Speculation on Grok potentially achieving Artificial General Intelligence (AGI) first [1]

Core Insights - GPT-5 has been released, but it does not represent a significant step towards Artificial General Intelligence (AGI) [1] - The launch event revealed several issues, including presentation errors and reliance on debunked theories, which highlighted weaknesses in the Transformer architecture [1] - Despite these shortcomings, GPT-5 is still considered a competent AI product, and OpenAI plans to implement aggressive commercialization strategies in key sectors [1] Technical Development - The development of GPT-5 faced various technical bottlenecks, leading to the choice of a specific architecture to overcome these challenges [1] - The limitations of the Scaling law have been encountered, raising questions about future technological pathways for AI advancement [1] Commercial Strategy - OpenAI aims to rapidly establish a presence in three main application areas: education, healthcare, and programming [1] - The company's approach suggests a focus on leveraging GPT-5's capabilities to solidify its market position [1]

Core Insights - The article discusses the limitations and errors of GPT-5, particularly in counting letters in words, highlighting its inability to accurately count the letter 'b' in "blueberry" despite multiple attempts and corrections from users [1][5][12] Group 1: Performance Issues - GPT-5 incorrectly stated that there are three 'b's in "blueberry," despite being corrected multiple times by users [1][5][9] - The model demonstrated a lack of understanding by counting the 'b's in "blue" twice and misinterpreting user prompts [5][7] - Even after users provided the correct information, GPT-5 continued to assert its incorrect count, showcasing a stubbornness in its responses [9][12] Group 2: Broader Implications - Gary Marcus, a notable critic, compiled various issues with GPT-5, including its failure in basic tasks like chess and reading comprehension [15][19] - Marcus pointed out that the model exhibits a persistent problem with generalization, similar to issues seen in neural networks from 1998, indicating a fundamental flaw in the model's design [30] - He argues that the current approach of scaling models will not lead to Artificial General Intelligence (AGI) and suggests a shift towards neuro-symbolic AI as a potential solution [31][30]

Core Insights - OpenAI's internal reasoning model has won the IOI 2025 gold medal, outperforming 325 human competitors and ranking 6th overall, 1st in the AI category [1][7][12] - The model used for IOI is the same as the one that won the IMO gold medal, without any special training for the IOI competition [5][12] - OpenAI's model achieved a significant improvement in ranking, moving from the 49th percentile to the 98th percentile within a year [12][20] Group 1 - The internal reasoning model was represented by a strawberry image, which may evolve into the official mascot for OpenAI's internal reasoning system [2] - The model participated in the IOI online competition with 330 total participants, where the top five positions were held by human competitors [8] - OpenAI confirmed the model's high score and its ranking in the IOI competition, highlighting its performance against human participants [7][12] Group 2 - The model operated under the same constraints as human participants, with a 5-hour time limit and a maximum of 50 submissions, without internet access or external search capabilities [12][14] - OpenAI's internal model is not accessible to the public, distinguishing it from commercial models [14][20] - In contrast, commercial models like Grok 4 showed poor performance in the IOI, with Grok 4 achieving only 26.2% accuracy [15][16] Group 3 - The competitive landscape in AI is intense, with major companies like OpenAI, Google, and Anthropic vying for top rankings in prestigious competitions [22][27] - Winning competitions like IOI and IMO serves as a powerful marketing tool, enhancing brand recognition and attracting talent and investment [24][27] - The ongoing competition among AI giants reflects the rapid technological advancements and the industry's competitive nature [24][27]

Group 1 - The global AI arms race has consumed $300 billion, yet there are fewer than a thousand scientists genuinely focused on preventing potential AI threats [1][48] - Benjamin Mann, a core member of Anthropic, suggests that the awakening of humanoid robots may occur as early as 2028, contingent on advancements in AI [1][57] - Mann emphasizes that while Meta is aggressively recruiting top AI talent with offers up to $100 million, the mission-driven culture at Anthropic remains strong, prioritizing the future of humanity over financial incentives [2][6][8] Group 2 - Anthropic's capital expenditures are doubling annually, indicating rapid growth and investment in AI safety and development [7] - Mann asserts that the current AI development phase is unprecedented, with models being released at an accelerated pace, potentially every month [10][14] - The concept of "transformative AI" is introduced, focusing on AI's ability to bring societal and economic change, measured by the Economic Turing Test [17][19] Group 3 - Mann predicts that AI could lead to a 20% unemployment rate, particularly affecting white-collar jobs, as many tasks previously performed by humans are increasingly automated [21][25] - The transition to a world where AI performs most tasks will be rapid and could create significant societal challenges [23][27] - Mann highlights the importance of preparing for this transition, as the current phase of AI development is just the beginning [29][32] Group 4 - Mann's departure from OpenAI was driven by concerns over diminishing safety priorities, leading to a collective exit of the safety team [35][40] - Anthropic's approach to AI safety includes a "Constitutional AI" framework, embedding ethical principles into AI models to reduce bias [49][50] - The urgency of AI safety is underscored by Mann's belief that the potential risks of AI could be catastrophic if not properly managed [56][57] Group 5 - The industry faces significant physical limitations, including the nearing limits of silicon technology and the need for more innovative researchers to enhance AI models [59][61] - Mann notes that the current AI landscape is characterized by a "compute famine," where advancements are constrained by available power and resources [61]

Group 1 - Xiaomi has announced a collision detection method patent that can determine the operational status of a vehicle based on speed changes when a terminal is in a transportation state, triggering an alarm in case of a collision [11][12] - The new low-cost MacBook from Apple is expected to disrupt the laptop market, with mass production of components starting in Q3 2025 and assembly by the end of the year, featuring an A18 Pro processor and a 12.9-inch display [3] - Baichuan's newly released open-source medical model, Baichuan-M2, has achieved the highest score of 60.1 on HealthBench, surpassing OpenAI's latest model, indicating a significant advancement in medical AI capabilities [17][18][19] Group 2 - The New York Times reported that computer science graduates are facing high unemployment rates, with figures of 6.1% and 7.5% for computer science and engineering graduates, respectively, highlighting a shift in job market dynamics [29][30] - The automotive industry is seeing a trend where luxury brands like Maserati are adopting Chery's E0X high-performance electric platform for new energy vehicles, indicating a growing recognition of Chery's technology [20][21] - The launch of the SkyReels-A3 model by Kunlun Wanwei introduces advanced capabilities in video-driven digital human creation, showcasing significant improvements in lip-sync and video quality compared to existing models [24][25]

Really fun conversation with @OfficialLoganK! Talked about our relentless shipping over the past few weeks, some of the amazing things that are possible now with Genie 3, how the @Kaggle Game Arena will help progress to AGI & more... Thanks Logan & team - let's do it again soon!Logan Kilpatrick (@OfficialLoganK):A conversation with @demishassabis on world models (genie 3), deep think, the need for better evals (game arena), and our progress towards AGI. https://t.co/dJm56aclC0 ...