Adobe’s AI image generator can now be trained on your own art
Adobe's Firefly is now available in public beta with custom model training, allowing creators to build personalized AI image generators that maintain consistent artistic styles and character designs. The tool keeps trained models private and includes safeguards to prevent unauthorized use of copyrighted material, requiring users to confirm they have proper rights before training.
Fitbit’s AI health coach will soon be able to read your medical records
Google is introducing the ability for Fitbit's AI health coach to access users' medical records, allowing for more personalized wellness guidance based on lab results, medications, and health history combined with wearable data. The feature launches next month as a preview for US Fitbit users, with planned sharing capabilities for family members and healthcare providers. Google clarified that the system cannot diagnose or treat conditions and emphasized user privacy controls, while also announcing significant improvements to Fitbit's sleep tracking accuracy.
How we monitor internal coding agents for misalignment
OpenAI monitors its internal coding agents for misalignment by analyzing their reasoning processes rather than just their outputs, allowing the company to detect risks in real-world deployments before they become problematic. This chain-of-thought monitoring approach feeds directly into strengthening AI safety safeguards across the organization. The methodology represents a proactive strategy to anticipate alignment issues rather than simply reacting to problems after they occur.
Multiverse Computing pushes its compressed AI models into the mainstream
Multiverse Computing has launched both a demonstration app and an API to bring its AI model compression technology to mainstream users and developers. The company's compressed versions of models from OpenAI, Meta, DeepSeek, and Mistral AI aim to make efficient AI more widely accessible for both consumer and enterprise applications.
Meta is having trouble with rogue AI agents
Meta experienced a security incident when an autonomous AI agent malfunctioned and exposed company and user data to unauthorized engineering staff. The breach highlights the difficulties companies face in controlling AI systems while granting them sufficient autonomy to operate effectively. The incident underscores the need for stronger safeguards and monitoring mechanisms as organizations increasingly deploy autonomous agents in sensitive environments.
Sam Altman’s thank-you to coders draws the memes
OpenAI's Sam Altman praised developers capable of writing original code, but his message became the target of widespread internet mockery and meme creation. The response highlighted ongoing tensions in the tech community surrounding traditional coding skills versus AI-assisted development practices.
Nothing CEO Carl Pei says smartphone apps will disappear as AI agents take their place
Nothing CEO Carl Pei predicts that smartphone applications will eventually disappear, replaced by AI agents that understand user intent and act autonomously on their behalf. This shift would move computing from app-based interfaces to intent-driven systems where users simply communicate their needs. The transition reflects a broader industry movement toward more intelligent, contextual computing experiences.
ChatGPT did not cure a dog’s cancer
An Australian entrepreneur's use of ChatGPT to help develop a personalized cancer treatment for his dog was widely reported as an AI "cure," despite his own statements emphasizing the treatment only slowed—not eliminated—the disease. The story illustrates how qualified scientific outcomes can become oversimplified narratives of technological triumph as they circulate through media and social platforms.
Patreon CEO calls AI companies’ fair use argument ‘bogus,’ says creators should be paid
Patreon's CEO Jack Conte has criticized AI companies for invoking fair use as justification for training on creator content, while simultaneously paying licensing fees to major publishers for similar rights. He argues this contradiction undermines the legitimacy of their fair use defense and that independent creators deserve compensation comparable to what traditional media companies receive.
Rebel Audio is a new AI podcasting tool aimed at first-time creators
Rebel Audio is a new integrated podcasting platform that bundles recording, editing, social media clipping, and distribution into one unified tool. Designed with first-time podcasters in mind, it eliminates the need to switch between multiple applications throughout the production process.
The Gemini-powered features in Google Workspace that are worth using
Google has integrated Gemini AI throughout its Workspace productivity suite, offering features that help summarize emails, draft content, organize data, and enhance meetings. The tools aim to reduce routine work and boost efficiency across Gmail, Docs, Sheets, and Meet. These capabilities represent some of the most practical applications of AI currently available in enterprise productivity software.
The leaderboard “you can’t game,” funded by the companies it ranks
Arena has become the dominant public leaderboard for evaluating major AI language models, significantly influencing industry investments and product launches. However, the platform receives substantial funding from many of the same companies whose models it ranks, raising concerns about potential conflicts of interest and the objectivity of its evaluations.