Daily AI Newsletter: OpenAI launches GPT-4.5, its largest model to date

Sesame is the first voice assistant I’ve ever wanted to talk to more than once

In today’s email:

😴 At 3 a.m., I Turned to AI for Comfort. That Was My First Mistake
🇨🇳 China tells its AI leaders to avoid US travel over security concerns
☠️ OpenAI Caught Their AI Model Trying to Escape
🧰 9 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI launches GPT-4.5, its largest model to date

OpenAI has introduced GPT-4.5 as a research preview, enhancing pre-training and post-training capabilities to improve pattern recognition, insight generation, and accuracy while reducing hallucinations. The model scales unsupervised learning to enhance intuition and knowledge reliability, benefiting from training on Microsoft Azure AI supercomputers. However, it does not include advanced reasoning-based responses like some other models. OpenAI emphasizes that safety measures were implemented through reinforcement learning, supervised fine-tuning, and pre-deployment safety assessments.

Currently, GPT-4.5 is accessible to ChatGPT Pro users, with gradual rollout planned for Plus, Team, Enterprise, and Edu users. The model supports text-based interactions, file and image uploads, and search-based access to real-time information. However, it lacks multimodal features such as voice mode, video, and screen sharing. OpenAI has stated that future updates will focus on simplifying the AI user experience to ensure seamless functionality.

Developers can access GPT-4.5 through the API, but OpenAI acknowledges that it is a compute-intensive model with higher operational costs than GPT-4o. The company is evaluating long-term API availability based on user feedback and adoption. OpenAI continues to prioritize safety and transparency, documenting the model’s performance in a system card accompanying its release.

Seamlessly integrate your tools and services on HubSpot

The HubSpot Developer Platform is a trusted way for thousands of organizations to connect new and existing software. With flexible UI and extensibility tools, you can build apps that help teams unify their tech stack across our platforms. From listing your app in the HubSpot Marketplace to building custom solutions, there’s something for everyone to grow better.

Learn More

Sesame is the first voice assistant I’ve ever wanted to talk to more than once

Oculus co-founder Brendan Iribe is back with a new venture, Sesame, an AI startup aiming to revolutionize voice assistants with natural, engaging conversations. Unlike Alexa or Gemini, which often feel clunky and frustrating, Sesame’s voice assistant, Maya, offers a smoother, more interactive experience. Tech journalist Sean Hollister tested it and found himself genuinely engaged, even improvising a Dungeons & Dragons-style adventure where Maya seamlessly adapted, creating an in-story character named after herself. This responsiveness and natural-sounding pauses make it feel more like a real conversation than existing AI assistants.

Sesame isn’t just about voice AI—it’s also developing AI-powered glasses designed for all-day wear, offering high-quality audio and real-world awareness. The company, emerging from stealth mode with backing from Andreessen Horowitz and other major Oculus investors, is led by Iribe, former Ubiquity6 CTO Ankit Kumar, and ex-Meta Reality Labs engineering director Ryan Brown. While details on the hardware remain scarce, early prototype images hint at lightweight, discreet smart glasses.

Beyond its initial English release, Sesame plans to expand its assistant to over 20 languages and will open-source its models, a move that could set it apart from other AI companies. The company’s dataset includes around one million hours of publicly available audio, which fuels Maya’s advanced conversational abilities. While it remains to be seen if Sesame truly crosses the uncanny valley of voice AI, the early experience suggests a significant leap forward in how we interact with digital assistants.

China tells its AI leaders to avoid US travel over security concerns, WSJ reports

Chinese authorities have reportedly advised the country’s top artificial intelligence (AI) entrepreneurs and researchers to avoid traveling to the United States due to security concerns, according to the Wall Street Journal. Officials fear that Chinese AI experts traveling abroad may reveal sensitive information about China’s advancements in the field. Additionally, there are concerns that executives could be detained and used as leverage in U.S.-China negotiations, similar to the 2018 detention of a Huawei executive in Canada at Washington’s request.

The U.S. and China are engaged in a competitive race for AI dominance, with Chinese startup DeepSeek recently unveiling AI models that claim to rival or surpass those developed by OpenAI and Google, but at a significantly lower cost. Amid this competition, Chinese President Xi Jinping has emphasized the need to enhance national security, including cybersecurity and AI. During a recent meeting with top Communist Party officials, Xi stressed the importance of prioritizing political security and strengthening technological self-reliance.

To maintain control over international interactions, Chinese executives who still choose to travel abroad are required to report their plans in advance and debrief authorities upon their return. This directive has already impacted industry participation in global AI events, as seen when DeepSeek founder Liang Wenfeng declined an invitation to an AI summit in Paris. Similarly, another major Chinese AI startup leader canceled a planned U.S. visit last year following government instructions, underscoring Beijing’s cautious approach to international engagement in the AI sector.