Sogou Input Method, truly a "dog" after all
🏛️ French antitrust regulator accuses NVIDIA
🌟 Runway Gen-3 opens to the public
🤝 Character.AI explores deals with Google and Meta
🔍 Tencent Yuanbao launches deep search
🐶 Sogou Input Method rolls out AI pet
🇨🇳 China's "AI construction guidelines" arrive
💰 ZhenFund donates to open-source project vLLM
📈 Abnormal Security valued at nearly $5 billion
💼 $2 billion new fund run by AI-driven trading
🧠 Fei-Fei Li: Current LLMs have no sentience
French Antitrust Regulator Accuses NVIDIA
France's antitrust regulator plans to bring anticompetitive conduct charges against NVIDIA, the AI chip giant. The move likely stems from a raid on the GPU industry last year. Under French antitrust law, violators face fines of up to 10% of their global annual revenue — for NVIDIA, that could mean as much as $6.09 billion. The regulator is concerned about market dependency on NVIDIA's CUDA chip programming software, and whether its investment in AI cloud provider CoreWeave risks reinforcing market dominance. Meanwhile, the United States and European Union are also scrutinizing NVIDIA's business, as global regulators focus on its market position.
Despite the charges, NVIDIA's stock has risen more than 151% this year, reaching a market cap of $3.06 trillion and making it the world's third most valuable company. CEO Jensen Huang emphasized at the annual shareholder meeting the company's pivot from gaming to data center operations, with plans to expand into new AI markets.
Runway Gen-3 Opens to the Public
Runway's [1] text-to-video model Gen-3 Alpha is now fully available, with access starting at $12 per month. While Gen-3 currently generates only 11-second 720P clips without background music, users widely report that it outperforms Sora and could transform the text-to-video industry. Users can log in through Runway's website, enter a prompt, and generate video. Gen-3 features physics simulation, with generated content strictly adhering to real-world characteristics, and has already been used to create short films, promotional ads, and other video content.
Though some users have expressed dissatisfaction with Runway's pricing, Gen-3's advanced features — including Motion Brush, camera controls, and Director Mode — will help users produce higher-quality videos. Runway's Gen-2 and other products remain free to use.
Character.AI Wants to Sell to Google and Meta
Character.AI [2], a startup founded by two former Google AI pioneers, captured attention a year ago with AI chatbots offering a novel form of entertainment: hyper-realistic roleplay with anime characters, TV celebrities, and historical figures. The company attracted millions of users to its website and app. Four months ago, Andreessen Horowitz and other investors injected $150 million at a $1 billion valuation.
But Character.AI's early lead has begun to erode. Now, the company founded in November 2021 may, like other AI startups, consider partnering with or being acquired by a larger tech company. Character.AI is currently exploring deals with rivals including Google and Meta — a move that could alter its trajectory as an independent firm.
Suno iOS App Launches
Suno, a rising name in AI-powered music creation, has released its iOS app [3].
The Suno app offers multiple subscription tiers. Users can start with a free plan offering 50 credits daily, or upgrade to Pro and Premier plans with more robust music production resources. The Pro plan costs $10 per month, while Premier runs $30 per month. Annual subscriptions are also available with a 20% discount.
Tencent Yuanbao Launches AI Deep Search
Tencent Yuanbao has rolled out an AI deep search mode: one-click generation of mind maps, tables, and event summaries.
In deep mode, Yuanbao expands queries to deliver more structured, comprehensive answers from both depth and breadth perspectives, while simultaneously generating content outlines, mind maps, and related figure-event breakdowns to give users a panoramic view of their search topic.
Tencent notes that search and Q&A account for as much as 45% of demand in LLM-related products. This upgraded deep mode will further expand question coverage and associative capability. For example, when users ask why gold prices are so high this year, Yuanbao can analyze from multiple dimensions, drawing on Tencent's ecosystem and other quality content sources.
Sogou Input Method Launches AI Pet
Tencent's Sogou Input Method has received a major upgrade, with new versions integrating AI-assisted writing, AI conversation, and AI pet features.
The AI writing assistant covers more than 130 use cases spanning social, creative, review and comment, and workplace scenarios. The new AI conversation feature incorporates 20 specialized AI assistants, covering life tips, dating advice, mock interviewers, and college major analysis. To balance productivity with leisure, Sogou Input Method has also introduced an AI pet function. Users can adopt their own cyber pet, interacting and conversing with it for a fun, warm companionship experience.
Sogou Input Method — truly a "dog" after all.
Image generated by DALL·E.
China's "AI Construction Guidelines" Arrive
China's Ministry of Industry and Information Technology, the Central Cyberspace Affairs Commission, the National Development and Reform Commission, and the National Standardization Administration have jointly issued the National Artificial Intelligence Industry Comprehensive Standardization System Construction Guidelines (2024 Edition) [4] (hereinafter referred to as the "Construction Guidelines").
By 2026, the linkage between standards and industrial technological innovation will continue to strengthen, with more than 50 new national and industry standards to be formulated, accelerating the formation of a standard system that guides high-quality AI industry development. Over 1,000 enterprises will carry out standard publicity, implementation, and promotion, with the effectiveness of standards serving enterprise innovation becoming more pronounced. Participation in formulating more than 20 international standards will advance the globalization of China's AI industry. Key construction areas cover foundational commonality, technical support, core technologies, product services, new industrialization, industry applications, and security governance.
ZhenFund Donates to Open-Source Project vLLM
ZhenFund recently announced a donation to the open-source AI project vLLM [5] to support its development. vLLM was developed by a UC Berkeley team to build the fastest, most user-friendly open-source large language model inference serving engine. Compared to Hugging Face Transformers, vLLM delivers up to 24x higher throughput without requiring model architecture changes. vLLM currently supports more than 30 cutting-edge open-source LLMs including Mistral, Llama, and Gemma.
A notable feature of vLLM is its broad hardware compatibility, supporting AMD GPUs, Intel CPUs/GPUs, Habana Gaudi, Google TPUs, AWS, and more. This enables faster, lower-cost inference services. Since going open source in June 2022, vLLM has surpassed 21.8k stars on GitHub, attracting over 300 active contributors from around the world.
Abnormal Security Valuation to Reach $5 Billion
AI startup Abnormal Security [6] is set to reach a $5 billion valuation, according to the latest reports.
Public records show Abnormal Security was founded in 2018 and uses AI to detect anomalous behavior and protect against cybersecurity threats including business email compromise (BEC). The company previously raised $210 million in Series C funding led by Insight Partners, with Greylock Partners and Menlo Ventures participating, at a $4 billion valuation.
$2 Billion New Fund Run by AI-Driven Trading
Hedge fund giant Bridgewater Associates has launched a new fund — one run and managed by machine learning. The fund has raised approximately $2 billion from fewer than 10 clients. The new investment vehicle, led by co-CIO Greg Jensen, began trading on Monday, July 1. Jensen has been at Bridgewater since 1996 and has been exploring how to integrate machine learning into the hedge fund's investments since 2012. He revealed that he invested in OpenAI's first funding round roughly a decade ago, and backed Anthropic a few years later.
According to people familiar with the matter, the new fund will also expand to incorporate models developed by companies including OpenAI, Anthropic, and Perplexity. The launch represents the latest step in the firm's transformation, which has also included major management changes. The new fund could reshape Bridgewater's hiring strategy and workforce composition, bringing in more data scientists.
Fei-Fei Li: Today's LLMs Have No Sentience
Stanford science professor Fei-Fei Li and philosophy professor John Etchemendy published an article in Time magazine [7] clarifying that current large language models do not possess "sentience" comparable to humans. They note that all human sensory experiences — hunger, pain, visual perception — arise from physiological states, which LLMs lack. Li further emphasizes that even future iterations of ChatGPT or other LLMs with larger parameter scales are unlikely to achieve genuine sentience.
In the article, the two professors counter arguments that LLMs' ability to report subjective experiences implies consciousness. They argue this analogy is inappropriate because statements generated by LLMs are merely probabilistic completions based on algorithms, not grounded in any actual physiological or mental states. Li and Etchemendy call for a deeper understanding of how sentience arises in biological organisms to truly achieve sentient AI.
Life is about experience, not about summaries and abstractions. As a creature of bounded rationality, I do not fantasize about fully, correctly, and objectively understanding the world. I am an adaptive system; survival and success depend on maintaining reasonably accurate images of the people and things around me. Because my world-image can only roughly approximate reality, I cannot seek optimization in anything — at best I can aim for satisfaction. Seeking the best is merely a waste of scarce cognitive resources.
— Herbert A. Simon, AI pioneer, Nobel laureate in Economics, Turing Award winner
Editor: Yuki Design: Ivan
For business partnerships, add on WeChat: Rwkfbcianvd
References
[1] Runway: https://app.runwayml.com/
[2] Character.AI: http://character.ai/
[3] Suno iOS app launch: https://apps.apple.com/us/app/suno-make-and-explore-music/id6480136315
[4] National Artificial Intelligence Industry Comprehensive Standardization System Construction Guidelines (2024 Edition): https://image.cnstock.com/attachment/20240702ba17187464e748d48a668ee8368ea8d6.pdf
[5] vLLM: https://github.com/vllm-project/vllm
[6] Abnormal Security: https://abnormalsecurity.com/
[7] Stanford science professor Fei-Fei Li and philosophy professor John Etchemendy in Time magazine: https://time.com/collection/time100-voices/6980134/ai-llm-not-sentient/