317
```markdown
layout: redirect
title: “Last Week in AI #317”
excerpt: “White House’s AI Action Plan 🇺🇸, OpenAI’s GPT-5 launch 🚀, Google embraces EU’s AI guidelines 🇪🇺, OpenAI’s Study Mode in ChatGPT 🎓, and more!”
image:
feature: assets/img/digests/317/President%20Donald%20Trump%20White%20House%20Website-211799041.jpg.webp?itok=a56S0EFq
credit:
---
layout: redirect
title: "Last Week in AI #317"
excerpt: "White House's AI Action Plan 🇺🇸, OpenAI's GPT-5 launch 🚀, Google embraces EU's AI guidelines 🇪🇺, OpenAI's Study Mode in ChatGPT 🎓, and more!"
image:
feature: assets/img/digests/317/President%20Donald%20Trump%20White%20House%20Website-211799041.jpg.webp?itok=a56S0EFq
credit: <a href="<Image Source Link>"> <Author> / <Source Name> </a>
categories: [digests]
permalink: /digests/the-three-hundred-and-seventeenth
sidebartoc: true
redirect: https://lastweekin.ai/p/317
---
### Top News
#### [White House Releases AI Action Plan to Cement U.S. Leadership](https://natlawreview.com/article/white-house-reveals-ai-action-plan)
On July 23, 2025, the White House introduced "Winning the AI Race: America’s AI Action Plan," a detailed strategy to solidify U.S. leadership in artificial intelligence (AI). This plan, following President Trump's executive order on Removing Barriers to American Leadership in AI, outlines over 90 federal policy actions across three pillars: Accelerating Innovation, Building American AI Infrastructure, and Leading in International Diplomacy and Security. The initiative aims to usher in a new era of human flourishing, economic competitiveness, and national security for the American people.
Key policies in the AI Action Plan include exporting American AI through partnerships between the Commerce and State Departments and industry, promoting rapid buildout of data centers, enabling innovation and adoption by removing federal regulations that hinder AI development, and upholding free speech in frontier models by updating federal procurement guidelines. The plan emphasizes the importance of AI as a cornerstone of American innovation and global influence, aiming to ensure that American workers and families thrive in the AI era. The White House Office of Science and Technology Policy Director Michael Kratsios, AI and Crypto Czar David Sacks, and Secretary of State and Acting National Security Advisor Marco Rubio all stressed the importance of winning the AI race for the U.S.'s economic and national security.
More on this:
* [White House Releases America’s AI Action Plan and Accompanying AI Executive Orders](https://www.quarles.com/newsroom/publications/white-house-releases-americas-ai-action-plan-and-accompanying-ai-executive-orders)
#### [OpenAI prepares to launch GPT-5 in August](https://www.theverge.com/notepad-microsoft-newsletter/712950/openai-gpt-5-model-release-date-notepad)
OpenAI is gearing up to launch its next-generation model, GPT-5, as early as August, following some delays. OpenAI CEO Sam Altman describes the model as a system that integrates a lot of their technology, including the o3 reasoning capabilities. The launch will also feature mini and nano versions of GPT-5, available through OpenAI's API. This unification of the o-series and GPT-series models is part of OpenAI's ongoing efforts to simplify and combine its large language models to create a more capable system that can eventually be declared artificial general intelligence (AGI).
The declaration of AGI is significant for OpenAI as it would force Microsoft to relinquish its rights to OpenAI revenue and its future AI models. However, it is unlikely that GPT-5 will meet the AGI threshold linked to OpenAI's profits. Altman has previously stated that GPT-5 won't have a "gold level of capability for many months" after launch. OpenAI is also preparing to launch an open language model, described as similar to o3 mini, which is expected to be released before GPT-5. This new model will be the first time that OpenAI has released an open-weight model since its release of GPT-2 in 2019.
#### [Big Tech split? Google to sign EU’s AI guidelines despite Meta snub](https://www.cnbc.com/2025/07/30/big-tech-split-google-to-sign-eus-ai-guidelines-despite-meta-snub.html)
Google has announced its intention to sign the European Union's guidelines on artificial intelligence (AI), a move that contrasts with Meta's decision to reject the same guidelines due to concerns about stifling innovation. Google believes that adhering to these guidelines will facilitate European citizens' access to new AI tools as they become available. The company's endorsement comes after Meta's refusal to sign the code, citing potential constraints on European AI innovation.
The European Commission, the executive body of the EU, has published a final version of its code of practice for general-purpose AI models, leaving the decision to sign up to individual companies. Google's President of Global Affairs, Kent Walker, emphasized the importance of prompt and widespread deployment of AI, suggesting that embracing AI could boost Europe's economy by 1.4 trillion euros ($1.62 trillion) annually by 2034.
#### [OpenAI launches Study Mode in ChatGPT](https://techcrunch.com/2025/07/29/openai-launches-study-mode-in-chatgpt/)
OpenAI has introduced a new feature called Study Mode in its AI chatbot, ChatGPT, aimed at fostering critical thinking skills in students. Rather than providing direct answers, the feature prompts users with questions to test their understanding of the material. The feature is being rolled out to users on Free, Plus, Pro, and Team plans, with plans to introduce it to Edu subscribers in the coming weeks. This move addresses concerns that while ChatGPT can be a useful tutoring tool, it may also negatively impact students' critical thinking abilities.
Despite the potential benefits of Study Mode, there are limitations to its effectiveness. Students can easily switch back to the regular mode if they want direct answers, and currently, there are no tools for parents or administrators to lock students into Study Mode. OpenAI's VP of Education, Leah Belsky, has indicated that the company may consider implementing such controls in the future. The launch of Study Mode is part of OpenAI's broader efforts to improve learning through ChatGPT, with plans to publish more information on how students use generative AI in their education.
### Other News
#### Tools
[Qwen Releases Qwen3-Coder-480B-A35B-Instruct: Its Most Powerful Open Agentic Code Model Yet](https://www.marktechpost.com/2025/07/22/qwen-releases-qwen3-coder-480b-a35b-instruct-its-most-powerful-open-agentic-code-model-yet/). Qwen3-Coder-480B-A35B-Instruct features a Mixture-of-Experts architecture with 480 billion parameters, enabling it to handle extensive codebases and autonomously interact with developer tools, excelling in agentic coding tasks and outperforming other open models in benchmarks.
[NVIDIA AI Dev Team Releases Llama Nemotron Super v1.5: Setting New Standards in Reasoning and Agentic AI](https://www.marktechpost.com/2025/07/27/nvidia-ai-dev-team-releases-llama-nemotron-super-v1-5-setting-new-standards-in-reasoning-and-agentic-ai/). The Llama Nemotron Super v1.5 offers enhanced performance and usability for reasoning and agentic tasks, featuring improved accuracy, efficiency, and the ability to operate on a single GPU, making it accessible for both individual developers and large enterprises.
[Zhipu AI Just Released GLM-4.5 Series: Redefining Open-Source Agentic AI with Hybrid Reasoning](https://www.marktechpost.com/2025/07/28/zhipu-ai-just-released-glm-4-5-series-redefining-open-source-agentic-ai-with-hybrid-reasoning/). The GLM-4.5 series, including GLM-4.5 and GLM-4.5-Air, introduces a hybrid reasoning approach with dual modes for complex reasoning and quick responses, achieving high benchmark scores and offering open-source access for broad commercial use.
[Google’s AI Overviews have 2B monthly users, AI Mode 100M in the US and India](https://techcrunch.com/2025/07/23/googles-ai-overviews-have-2b-monthly-users-ai-mode-100m-in-the-us-and-india/). Google's AI Overviews, a feature providing AI-generated summaries of search results, has seen significant growth, now reaching 2 billion monthly users, while AI Mode, offering a chat-based search experience, has over 100 million users in the U.S. and India.
[Microsoft Edge transforms into an AI browser with new Copilot Mode](https://www.theverge.com/news/714435/microsoft-edge-copilot-mode-ai-features). The new Copilot Mode allows users to utilize AI for tasks such as searching across open tabs, booking reservations, and comparing products, with features like voice navigation and optional integration with browser history and credentials.
[GitHub Introduces Vibe Coding with Spark: Revolutionizing Intelligent App Development in a Flash](https://www.marktechpost.com/2025/07/24/github-introduces-vibe-coding-with-spark-revolutionizing-intelligent-app-development-in-a-flash/). Spark allows users to create full-stack apps from natural language prompts, and is available to Copilot Pro+ subscribers for $39 a month.
[Figma’s AI app building tool is now available for everyone](https://www.theverge.com/news/712995/figma-make-ai-general-availability-announcement). Figma Make, the prompt-to-app coding tool, allows users to build prototypes and apps using natural language and design references, with access varying based on subscription plans.
[Runway Launches New Aleph Model That Promises Next-Level AI Video Editing](https://www.cnet.com/tech/services-and-software/runway-launches-new-aleph-model-that-promises-next-level-ai-video-editing/). The new model, Aleph, focuses on "fluid editing" by allowing users to make complex video edits through simple prompts, maintaining continuity and consistency across frames.
[Photoshop just made it shockingly easy to edit objects and people into photos](https://www.theverge.com/news/715073/adobe-photoshop-ai-harmonize-composite-editing-feature). Photoshop's new generative AI features include the "Harmonize" tool, which automatically adjusts new objects to blend seamlessly into images, and improved tools for image upscaling and object removal, all while implementing safeguards like Content Credentials to ensure accountability.
[Google Gemini on Wear OS widely rolling out](https://9to5google.com/2025/07/23/gemini-wear-os-update-3/). Google Gemini on Wear OS replaces Google Assistant and offers enhanced natural-language query support, access to various apps and extensions, and the ability to provide responses in text, images, and voice.
[Google is testing a vibe-coding app called Opal](https://techcrunch.com/2025/07/25/google-is-testing-a-vibe-coding-app-called-opal/). The app allows users to create and share mini web apps using text prompts and a visual workflow, aiming to make app development accessible to non-technical users.
[Anthropic unveils new rate limits to curb Claude Code power users](https://techcrunch.com/2025/07/28/anthropic-unveils-new-rate-limits-to-curb-claude-code-power-users/). Anthropic is implementing these limits to address continuous background usage and account sharing, affecting less than 5% of subscribers, while offering additional usage options for Max plan users at standard API rates.
[Google’s NotebookLM can now make narrated slideshows with AI](https://www.theverge.com/news/715283/google-notebooklm-video-overviews-studio-ai). The new Video Overviews feature allows users to create narrated slideshows with AI-generated visuals, images, and diagrams, and is part of a broader update to NotebookLM's Studio tab, which now supports multiple outputs and a refreshed interface.
[Google Photos can now animate your photos into short videos](https://www.techradar.com/ai-platforms-assistants/gemini/google-photos-can-now-animate-your-photos-into-short-videos). Google Photos can now animate your photos into short videos. The new feature uses Google's Veo 2 AI video model to create six-second clips with subtle movements, and includes a Remix tool for restyling images, all while ensuring AI-generated content is clearly labeled with visible and invisible watermarks.
#### Business
[OpenAI expands Oracle data center deal, says parts of Stargate 1 in Texas are operational](https://www.cnbc.com/2025/07/22/openai-oracle-stargate-texas.html). The expansion will create over 100,000 jobs in the U.S. and increase the total Stargate AI data center capacity under development to over 5 gigawatts.
[Meta names Shengjia Zhao as chief scientist of AI superintelligence unit](https://techcrunch.com/2025/07/25/meta-names-shengjia-zhao-as-chief-scientist-of-ai-superintelligence-unit/). Zhao, a former OpenAI researcher known for his work on AI reasoning models, will lead Meta Superintelligence Labs' research agenda alongside Alexandr Wang, as Meta bolsters its AI capabilities with significant investments in talent and infrastructure.
[Nvidia AI chip challenger Groq said to be nearing new fundraising at $6B valuation](https://techcrunch.com/2025/07/29/nvidia-ai-chip-challenger-groq-said-to-be-nearing-new-fundraising-at-6b-valuation/). Groq is reportedly in discussions to secure $600 million in funding, which would double its valuation to nearly $6 billion, with the round led by Disruptive and follows recent partnerships with Bell Canada and Meta.
[A.I.-Driven Education: Founded in Texas and Coming to a School Near You](https://www.nytimes.com/2025/07/27/us/politics/ai-alpha-school-austin-texas.html). Alpha School plans to expand its AI-driven educational model, which personalizes learning based on individual student skills and interests, to multiple cities nationwide this fall.
[AI referrals to top websites were up 357% year-over-year in June, reaching 1.13B](https://techcrunch.com/2025/07/25/ai-referrals-to-top-websites-were-up-357-year-over-year-in-june-reaching-1-13b/). AI referrals, while still trailing behind Google Search in overall traffic, have shown significant growth, particularly in the news and media category, where referrals increased by 770% since June 2024, with ChatGPT being the dominant source.
[SK Hynix Quickens AI Investments After Posting Record Profit - Bloomberg](https://www.bloomberg.com/news/articles/2025-07-23/sk-hynix-reports-record-profit-and-sales-on-booming-ai-demand). None
#### Research
[TikTok Researchers Introduce SWE-Perf: The First Benchmark for Repository-Level Code Performance Optimization](https://www.marktechpost.com/2025/07/21/tiktok-researchers-introduce-swe-perf-the-first-benchmark-for-repository-level-code-performance-optimization/). SWE-Perf evaluates the ability of large language models to optimize code performance in real-world repositories, highlighting a significant gap between current model capabilities and human expert performance.
[AlphaGo Moment for Model Architecture Discovery](https://arxiv.org/abs/2507.18074). ASI-Arch achieves this by autonomously discovering state-of-the-art neural architectures, surpassing human limitations and uncovering new design principles.
[Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report](https://arxiv.org/abs/2507.16534). The report evaluates the risks associated with general-purpose AI models, categorizing them into green, yellow, and red zones based on their potential impact on public health, national security, and societal stability, with current models mostly falling within the green and yellow zones.
[Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains](https://arxiv.org/abs/2507.17746). Rubrics as Rewards (RaR) introduces a framework that uses structured rubrics as reward signals in reinforcement learning, enabling improved performance and alignment with human preferences across diverse domains, including medicine and science.
[Working with AI: Measuring the Occupational Implications of Generative AI](https://arxiv.org/abs/2507.07935). The study analyzes the use of Microsoft Bing Copilot in various work activities and computes an AI applicability score for different occupations, identifying high scores for knowledge and administrative roles.
[Inverse Scaling in Test-Time Compute](https://arxiv.org/abs/2507.14417). Increased test-time compute can lead to a decline in performance and worsen particular reasoning errors in large reasoning models when evaluated across varying reasoning lengths.
[The Serial Scaling Hypothesis](https://arxiv.org/abs/2507.12549). The Serial Scaling Hypothesis suggests that for complex machine learning tasks involving reasoning, planning, or interacting systems, increasing serial computation is essential, as parallel computation alone is insufficient for progress.
[Group Sequence Policy Optimization](https://arxiv.org/abs/2507.18071). Group Sequence Policy Optimization (GSPO) is a reinforcement learning algorithm that improves training efficiency and performance by using sequence-level importance ratios and clipping, which has notably enhanced the stability of Mixture-of-Experts RL training and simplified RL infrastructure design.
[Learning without training: The implicit dynamics of in-context learning](https://arxiv.org/abs/2507.16003). The article explores the concept of in-context learning in large language models, proposing that implicit weight updates occur during inference, transforming context into a weight update for neural networks, and providing an explicit formula for this process, which challenges previous assumptions by allowing for more generic prompts and focusing on the MLP layer rather than the self-attention layer.
[Step-Audio 2 Technical Report](https://arxiv.org/abs/2507.16632). Step-Audio 2 is an end-to-end large audio language model that processes raw audio inputs to generate both text and audio outputs, incorporating para-linguistic and non-vocal information, and utilizes retrieval-augmented generation and external tools for enhanced speech interaction and expressiveness.
[Yume: An Interactive World Generation Model](https://arxiv.org/abs/2507.17744). Yume allows users to explore and interact with dynamic virtual worlds generated from images using keyboard inputs, offering improved camera control and visual quality through innovations like Quantized Camera Motion and Anti-Artifact Mechanism.
#### Concerns
[GitHub Goes After Vibe Coding Fans with the Public Preview of GitHub Spark](https://www.hackster.io/news/github-goes-after-vibe-coding-fans-with-the-public-preview-of-github-spark-d54d80e3a6f4). Despite the potential benefits of AI chatbots, their influence on body image issues is raising concerns as they may inadvertently reinforce negative self-perceptions and exacerbate body dysmorphia.
#### Policy
[America’s AI watchdog is losing its bite](https://www.technologyreview.com/2025/07/24/1120645/americas-ai-watchdog-is-losing-its-bite/). The Trump administration's recent actions aim to reduce regulatory oversight by the FTC on AI companies, potentially allowing for faster AI deployment with fewer checks on accuracy, fairness, or consumer harm.
<hr>
Copyright © 2024 Skynet Today, All rights reserved.
