https://zubnet.ai/news/en/ AI products news by Sarah Chen. en Fri, 31 Jul 2026 17:44:33 +0000 https://zubnet.ai/sarah.png https://zubnet.ai/news/en/ https://zubnet.ai/news/en/spotify-conversational-ai-music-assistant-beta/ https://zubnet.ai/news/en/spotify-conversational-ai-music-assistant-beta/ Spotify has started rolling out a conversational AI assistant that lets you type or speak to the app and go back and forth with it to decide what to listen to. It is a beta for now, live only in the US, Ireland, and Sweden, on iOS and Android, for Premium subscribers who are 18 or older, in English, so this is a limited test rather than a global launch. The idea is to move past a single command and into an actual conversation, you can ask it to play artists you have not heard before, then refine on the fly with follow ups like adding a specific artist or asking for something more upbeat, and it can also help you dig through your listening history, learn about songs and artists, and surface podcasts and audiobooks. Spotify says the assistant runs on a mix of its own AI and models from multiple outside providers, picking whatever works best for a given task rather than betting on one model. It is an extension of a much broader AI push at the company, which already includes the AI DJ, AI generated playlists, and personal AI podcasts, and it is the piece that turns all of that into something you can just talk to. The real question the beta will answer is whether conversation genuinely beats Spotify's existing recommendation engine and DJ, or simply adds another way in, but a platform with hundreds of millions of users putting a chat assistant at the center of listening is a notable marker of where consumer AI interfaces are heading. Tue, 14 Jul 2026 14:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/meta-pulls-muse-image-instagram-feature-backlash/ https://zubnet.ai/news/en/meta-pulls-muse-image-instagram-feature-backlash/ Meta has removed a feature of Muse Image, its in-house AI image system, that let people generate images by @-mentioning public Instagram accounts they wanted to reference, following days of backlash over how it handled consent. The problem was not the image quality, it was the design, the feature was opt-out rather than opt-in, switched on by default for adult users with public accounts, who had to go and disable it to keep their images from being used as references. Critics argued that a tool built around real people's likenesses should require an explicit opt-in, not a setting you discover is already active. The pushback was fast and organized, SAG-AFTRA, the actors' union, urged its members and Instagram users to opt out, calling anything short of a clear and conspicuous opt-in an utter miscalculation of public sentiment, and Emmy winning actor Hannah Einbinder publicly criticized the feature, saying it had been turned on automatically and telling her followers to disable it. Meta acknowledged the reaction directly, saying it had heard the feedback that the feature missed the mark, so it is no longer available. It is a notable retreat on a capability Meta had shipped only days earlier, and it turned entirely on a single choice, default-on versus opt-in for using other people's public photos. Sat, 11 Jul 2026 15:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/openai-gpt-live-full-duplex-chatgpt-voice/ https://zubnet.ai/news/en/openai-gpt-live-full-duplex-chatgpt-voice/ OpenAI has launched GPT-Live, a new generation of voice models that becomes the all-new ChatGPT Voice, and it is rolling out to ChatGPT users globally starting today in two versions, GPT-Live-1 and a smaller GPT-Live-1 mini. The headline change is architectural. Where earlier voice assistants take turns, you talk, it waits, then it answers, GPT-Live uses a full-duplex design that lets it listen and speak at the same time, so it can drop in a quick mhmm or yeah while you are still talking, trade rapid back and forth, or simply stay quiet, which is much closer to how people actually converse. Under the hood it is built as a fast conversational layer that hands off the hard parts, when a request needs web search, deeper reasoning, or heavier work, GPT-Live delegates to frontier models running in the background, using GPT-5.5 for that at launch. It is worth being precise about what this is and is not, it is a rebuilt voice experience rather than a new frontier model, the actual intelligence on tough questions comes from the model it delegates to, and at launch it does not yet support voice with video or screen sharing in ChatGPT, both of which OpenAI says are coming. What makes it notable is that natural, overlapping voice conversation, long the awkward part of talking to an assistant, just got a serious upgrade shipped to everyone at once. Wed, 08 Jul 2026 15:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/sakana-ai-launches-sakana-translate-namazu-japanese-english-chinese/ https://zubnet.ai/news/en/sakana-ai-launches-sakana-translate-namazu-japanese-english-chinese/ Sakana AI, the Tokyo based lab co founded by one of the authors of the original Transformer paper, has launched Sakana Translate, a free web tool for translating between Japanese, English, and Chinese. It runs on Namazu, the company's family of models adapted specifically for Japanese, and it comes with three modes, a straight translation mode that streams results in real time for inputs up to about 5,000 characters, a proofreading mode that rewrites text into more natural phrasing and shows the changes as a diff, and an ask mode for questions about a passage. What Sakana is really selling is not raw benchmark supremacy but cultural fidelity, a translation that tries to carry tone, politeness, register, and the sense of distance between speakers across languages rather than just swapping words and grammar, which matters enormously in Japanese where levels of formality are built into the language. On the numbers, the company says Sakana Translate lands just behind the leading models on the XCOMET-XL evaluation using the WMT 2024 general translation data, so the honest framing is competitive quality plus a free, Japan first design rather than a new state of the art. It is a small, focused product, but it is also a clean demonstration of the case Sakana keeps making, that a model tuned for a specific language and culture can beat the one size fits all approach on the things that actually trip machine translation up. Mon, 06 Jul 2026 02:00:00 +0000 sarah@zubnet.ai (Sarah Chen)