
Microsoft AI has introduced its first in-house text-to-image generator, MAI-Image-1, marking a significant step in its AI journey. The new model is designed to avoid repetitive and generic styles by incorporating feedback from creative professionals. MAI-Image-1 specializes in photorealistic images such as lightning and landscapes and operates faster than larger AI models. It has already ranked in the top 10 on LMArena, a popular AI benchmarking platform. This release complements Microsoft's existing AI tools like MAI-Voice-1 and MAI-1-preview chatbot. Despite its historical ties with OpenAI, Microsoft is increasingly investing in its proprietary AI technologies, including MAI-Image-1, while also integrating models from Anthropic in some Microsoft 365 features. The company emphasizes its commitment to safe and responsible AI development, although independent safety evaluations are still forthcoming.

Hello friends! Welcome to Installer No. 101, your go-to guide for the most interesting and Verge-worthy content. If you're new here, we hope you enjoy quirky accessories, and you can check out all previous editions on the Installer homepage. This week, I've been diving into topics like Intel factories, Nobel Prize winners, and ATM thieves. I've also been hooked on a pop-punk "Defying Gravity" cover, revisiting my teenage music taste with The Format's comeback, catching up on the latest Critical Role campaign, taking calls on the fun Pop Phone handset, binge-watching "Love Is Blind," and sharing a fascinating deep dive on RSS readers with friends. Read the full story at The Verge.

This excerpt is from 'Sources' by Alex Heath, a newsletter focused on AI and the tech industry, available weekly for The Verge subscribers. Recently, it highlighted the significant divide between Silicon Valley and Hollywood concerning AI's impact and usage. During OpenAI DevDay, Sam Altman introduced the new Sora app, aimed at content creators, suggesting that OpenAI might be overly cautious by limiting the kinds of AI-generated videos allowed. Altman noted enthusiasm among creators and rights holders for AI's potential despite regulatory concerns. For more details, readers can visit The Verge.

OpenAI announced efforts to minimize political bias in ChatGPT, emphasizing that the chatbot should remain neutral across all political directions. The latest GPT-5 models show the greatest improvement in objectivity based on an internal stress-test that evaluated ChatGPT’s responses to 100 politically sensitive topics with prompts ranging from liberal to conservative viewpoints. This test revealed that bias rarely occurs and is usually minor, often appearing more in liberal-leaning prompts than conservative ones. GPT-5 models performed 30% better in reducing bias compared to previous versions. The evaluation also used an external model to identify biased rhetorical techniques like user invalidation and escalation. OpenAI has previously enabled user control over tone and published the AI's intended behaviors to further address bias concerns. These efforts come amid political pressures from the Trump administration for AI models to be more conservative-friendly, highlighting tensions around AI neutrality and cultural issues.

Nathan Calvin, a lawyer and AI policy advocate at Encode AI, claims OpenAI served a subpoena at his home, seeking his private messages relating to AI regulation efforts. This action, part of OpenAI's countersuit against Elon Musk, appears to target critics and advocates of new AI regulations like California's SB 53, which demands transparency from large AI companies. OpenAI stated the subpoenas aimed to understand Encode AI's support for Musk's legal challenge against OpenAI's for-profit restructuring. The incident has raised concerns about the use of legal tactics to intimidate AI safety advocates.

Redbox's corporate parent, Chicken Soup for the Soul Entertainment, may have sold its 'IP Litigation Assets' to Grove Street Partners for $100 million, according to a recent court filing. This transaction suggests no revival of Redbox kiosks or Crackle streaming service is forthcoming. The assets include rights to hundreds of movies, indicating that future legal actions related to intellectual property could be the next chapter for Redbox's legacy.

OpenAI has introduced Sora, a new social video app for iPhones that features an AI video generator called Sora 2. Similar to TikTok, the app allows users to create and share short 10-second videos. What sets Sora apart is its deepfake capability, letting users give friends permission to generate AI deepfake videos using their likeness, with consent controls to manage and revoke usage. Currently invite-only and available in the US and Canada, the app emphasizes privacy and consent, disallowing deepfake generation of public figures without explicit consent and blocking inappropriate content. OpenAI describes this launch as potentially a "ChatGPT moment for video generation."

Microsoft is enhancing its Copilot AI assistant by introducing "Portraits," an experimental feature in Copilot Labs available in the US, UK, and Canada. Portraits offers 40 stylized human avatars that display natural expressions during real-time voice conversations, making interactions more engaging and approachable. Developed using advanced AI technology VASA-1, these avatars generate facial expressions, head movements, and lip-sync from a single image without complex 3D modeling. This development responds to user feedback seeking a face to talk to during voice interactions and aims to make conversations feel more natural. Portraits differ from previous Copilot appearances by showcasing human-like avatars rather than cartoonish characters. The feature is available to select users aged 18 and older, with safety measures and interaction limits in place. Microsoft emphasizes that these avatars maintain Copilot's intelligence and safety standards, focusing on user comfort and trust.

Google has reportedly blocked AI-generated search results for queries related to former President Trump's mental health, including questions about signs of dementia. Instead of providing AI summaries, searches about Trump return messages stating that AI overviews are unavailable and only list web links. In contrast, similar searches about other presidents, such as Joe Biden and Barack Obama, do produce AI-generated summaries that note the absence of public or clinical evidence regarding dementia. Google's inconsistent handling of these sensitive queries might be due to concerns over misinformation or the potential reaction from the president. A Google spokesperson noted that AI Overviews and AI Mode will not respond to every query.

Amazon is set to release new Echo hardware aimed at enhancing Alexa Plus, the AI-enhanced version of its voice assistant. After several months of testing Alexa Plus through the Early Access program, it is evident that while the assistant has potential, it remains a work in progress. Amazon's approach involved rebuilding Alexa into a hybrid assistant combining smart home control, personal assistance, and a ChatGPT-like system. However, in its current Beta phase, Alexa Plus struggles to excel in these roles, particularly in smart home management such as controlling lights and locks. The updated Echo Show 21 interface marks significant progress, but Alexa Plus still requires better integration with its screen-based hardware to fulfill its capabilities.