GPT-5 next month, while American and China swap action plans
Get bigger weekly updates! Free subscribers receive the top stories each week, while Paid subscribers will get a few extra stories. All support for Handy AI directly helps us maintain the newsletter and keep the information flowing.
last week’s top stories
🚀 OpenAI readies GPT-5 for August debut. GPT-5, OpenAI’s next-generation model, is reportedly set to launch in early August after months of testing and tweaks. CEO Sam Altman recently teased GPT-5’s capabilities (during a podcast, he let the unreleased model answer a complex question he couldn’t, calling it a “here it is moment” and admitting he felt “useless” next to the AI). The new model is expected to be not just more powerful at coding and reasoning, but to combine traditional GPT-style skills with the advanced logic of OpenAI’s “o-series” reasoning models. Rumors also suggest OpenAI will launch smaller GPT-5 Mini and Nano versions via API to broaden access to this cutting-edge AI. Read more
🏛️ White House unveils sweeping U.S. AI Action Plan. President Trump’s administration released a comprehensive AI strategy aiming to fast-track American AI development and “win” the global tech race. The blueprint, with ~90 directives, calls for loosening environmental rules to speed data center construction and expanding exports of U.S. AI hardware and software to allies. Trump signed multiple executive orders enacting parts of the plan, framing AI leadership as a 21st-century arms race that America “is going to win”.
🤝 China proposes global AI cooperation framework. China’s government discussed plans to create a new international organization for AI collaboration, positioning itself as an alternative leader to the U.S. in shaping AI’s future. Speaking at the World AI Conference in Shanghai, Premier Li Qiang advocated a shared global AI governance framework and offered to share China’s tech advances with developing nations. Read more
🧠 Meta poaches ex-OpenAI scientist for “Superintelligence” lab. Meta hired Shengjia Zhao, a co-creator of ChatGPT and former OpenAI research lead, to be chief scientist of its new “Superintelligence Labs” division. CEO Mark Zuckerberg announced the move, calling Zhao a “pioneer” who helped build GPT-4 and other frontier models. Read more
🎥 Runway unveils Aleph AI tool for video editing. Startup Runway introduced Aleph, a new AI model that can edit and generate video via text prompts. The system essentially acts as a post-production suite-in-a-model, automatically relighting scenes or transforming environments (e.g. turning a sunny park video into a rainy night) while keeping the rest of the shot intact. Read more
🧬 AI designs cancer-killing proteins in weeks. An AI platform developed by researchers at DTU and Scripps crafted custom immune proteins that guide T-cells to attack tumors in just 4–6 weeks. In lab tests, T-cells equipped with the AI-designed proteins recognized and destroyed cancer cells, including patient-derived melanoma targets, showing the potential for rapid personalized immunotherapies. Read more
☁️ OpenAI & Oracle expand Stargate AI supercomputers. OpenAI has inked a major deal with Oracle to build an additional 4.5 gigawatts of AI data center capacity in the U.S., extending their joint “Project Stargate” initiative. The expansion, part of an up to $500 B plan unveiled with SoftBank in January, will dramatically boost the cloud infrastructure available for training and running next-gen models like GPT-5. Read more
✂️ “DOGE” deploys AI to slash U.S. regulations. The White House’s Department of Government Efficiency (DOGE) is using an AI tool to review over 200,000 federal rules and identify which ones can be axed. Officials say the plan isn’t final, but praised DOGE’s team (initially led by Elon Musk) as “the best and brightest” pushing bureaucratic automation (notwithstanding an earlier DOGE AI tool that hallucinated some facts). Read more
🧪 AI Research of the Week
A Survey of Context Engineering for Large Language Models
From Chinese Academy of Sciences, UC Merced, University of Queensland, Peking University, and Tsinghua University
Jake's Take: This massive survey paper (analyzing over 1,400 research papers!) introduces "Context Engineering" as a formal field that goes way beyond basic prompt engineering. The researchers created a comprehensive roadmap breaking down how we feed information to AI models into foundational components (how we retrieve, process, and manage context) and system implementations (like RAG, memory systems, and multi-agent setups).
Their key finding is that there's a fundamental imbalance in current LLMs where they're incredibly good at understanding complex, long contexts you give them, but they struggle to generate equally sophisticated, lengthy outputs in return. The authors argue that fixing this input-output imbalance should be a top priority for researchers.
The paper essentially provides one of the first unified frameworks for understanding how context flows through AI systems, making it invaluable for anyone building context-aware applications.
and then, even more news…
✋ Meta’s prototype wristband reads minds (well, muscle signals). Meta researchers revealed a gesture-controlled wristband that lets people control computers using subtle nerve signals from their arm. The device uses surface electromyography (sEMG) to detect the electrical impulses that muscles generate, effectively sensing what action you intend to do before you even move. In testing with spinal injury patients, the AI-powered band enabled users with paralysis to move cursors and “air-write” messages, offering a noninvasive alternative to brain implants like Neuralink. Meta envisions this neural wristband as a future interface for AR glasses and other devices (no keyboard or touch required). Read more
💰 Tesla inks $16.5 B chip deal with Samsung. Elon Musk announced that Tesla has signed a $16.5 billion agreement for Samsung to manufacture its next-generation “AI 6” self-driving chips. The multi-year deal will utilize Samsung’s new Texas semiconductor fab to produce Tesla’s custom AI processors, which Musk said will supercharge the automaker’s autonomous vehicle and robotics efforts. Samsung’s stock jumped ~7% on the news, as the contract should boost its lagging contract-chip business in competition with TSMC. Musk noted Tesla engineers will work on-site to maximize yield at the Texas plant, and teased that the “$16.5B is just the bare minimum,” suggesting actual output (and revenue) could be several times higher as Tesla ramps production of AI-powered vehicles. Read more
🏥 OpenAI’s “AI co-pilot” cuts medical errors in Kenya. A large real-world study in Kenya showed an OpenAI-powered clinical assistant can significantly reduce diagnostic mistakes by primary care providers. In trials across 15 clinics (nearly 40,000 patient visits), clinicians using the AI Consult tool had 16% fewer misdiagnoses and 13% fewer treatment errors compared to a control group. The AI quietly monitors patient exams and nudges the clinician only if they’re potentially making a mistake, akin to a virtual expert over their shoulder. Doctors said the system not only caught errors but also served as a learning aid, boosting their confidence and expanding their medical knowledge over time. With this success at Penda Health clinics, OpenAI’s pilot hints at a future where AI “co-pilots” help frontline health workers improve care quality in resource-limited settings. Read more
⌚ Amazon acquires AI wearable startup “Bee”. Amazon is buying Bee, a San Francisco startup that makes an AI-powered wristband which records and transcribes your conversations. Bee’s $50 wearable (the “Pioneer” bracelet) uses a built-in mic and on-device AI to capture everything you say throughout the day and then generate summaries, to-do lists, or other notes from the audio logs. Amazon confirmed the deal after Bee’s CEO posted about it, indicating the gadget will join Amazon’s Devices group under exec Panos Panay. It’s a bold move into wearable AI for Amazon, which dabbled in Alexa glasses and Halo bands, and comes as OpenAI and others also explore “always-listening” AI assistants in new form factors. Read more
🎯 U.S. to arm Ukraine’s drones with 33,000 AI “strike” kits. The Pentagon is funding a deal with Auterion (a U.S. defense tech firm) to send 33,000 AI-powered guidance modules for Ukrainian drones by end of 2025. These $50 million worth of kits will upgrade manually-piloted FPV drones with the ability to autonomously track and hit targets up to 1 km away, using on-board computer vision to lock on in the final moments of flight. The AI guidance helps drones overcome heavy Russian jamming by not relying on constant radio control. Auterion’s CEO Lorenz Meier said they had already shipped thousands of earlier-gen “AI strike” systems, but this new order boosts support more than tenfold as Ukraine seeks to counter mass assaults with smarter, self-guided drones. Read more