Matthias Bastian
Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
Read full article about: OpenAI releases gpt-oss-safeguard open source models for flexible AI safety
OpenAI has launched gpt-oss-safeguard, a new set of open source models built for flexible security classification. The models come in two sizes, 120b and 20b, and are available under the Apache 2.0 license for anyone to use and modify. Unlike traditional classifiers that need to be retrained whenever safety rules change, these models can interpret policies in real time, according to OpenAI. This lets organizations update their rules instantly, without retraining the model.
The models are designed to be more transparent as well. Developers can see exactly how the models make decisions, making it easier to understand and audit how security is enforced. gpt-oss-safeguard is based on OpenAI's gpt-oss open source model and is part of a larger collaboration with ROOST, an open source platform focused on building tools and infrastructure for AI safety, security, and governance.
Read full article about: Bill Gates calls AI "the biggest technical thing ever in my lifetime," but warns of a bubble
Bill Gates recently compared the current wave of excitement around AI to the dot-com bubble, while making it clear this isn't just hype. In a CNBC interview, Gates said companies are pouring huge sums into chips and data centers, even though most haven't turned a profit from AI yet. He expects some of these bets will end up as costly failures. Still, Gates calls AI "the biggest technical thing ever in my lifetime," describing its economic potential as enormous. At the same time, he cautions that the surge in new data centers could drive up electricity costs.
Gates isn't alone in his concerns. Other industry leaders, including OpenAI CEO Sam Altman and AI researchers like Stuart Russell and Yann LeCun, have recently warned that the current AI boom could end with a crash if expectations get too far ahead of real progress.
Read full article about: Google adds Gemini AI app builder to AI Studio
Google has introduced a new AI app builder inside AI Studio, letting users create AI-powered applications with just a text prompt. This update puts Google on par with Microsoft, OpenAI, Anthropic, and others offering similar tools for building simple apps without coding.
One of the key features is deep integration with Google's multimodal Gemini models. Users can add image generation with the "Nano Banana" model or bring in Gemini’s voice synthesis capabilities directly into their apps. Design changes can be made visually on the app screen, with the option to guide edits using text instructions.
Developers can plug in their own API keys once they use up the free quota. The revamped app gallery offers templates and ideas to help users get started. For example, it took me just 15 minutes to build a simple "The Decoder" game that doesn't actually do anything.
Read full article about: Anthropic expands Claude's financial tools for analysts
Anthropic is adding new features to Claude designed for financial analysts, including an Excel extension, expanded data connections, and a set of AI-powered skills for tasks like cash flow modeling and valuation comparisons. The Claude for Excel add-in, powered by Sonnet 4.5, is now in beta for select business customers. It can analyze, edit, and comment on spreadsheets directly within Excel.
The release also features real-time data connections to providers including Moody's, LSEG, and Aiera. Six new agent functions are intended to support due diligence reviews and company analyses, further automating standard analytical tasks.
With these additions, Anthropic goes head-to-head with similar tools from Microsoft Copilot and ChatGPT. Still, none of these platforms have a well-defined error rate for everyday office work. Even though reasoning models have made progress on math-heavy tasks, these systems are still probability-based and can and will make mistakes.
Read full article about: OpenAI is building an AI model that can generate music from text or audio promptsOpenAI is building an AI model that can generate music from text or audio prompts
OpenAI is developing an AI model for music generation, putting it in direct competition with startups like Suno and Udio. The company is reportedly collaborating with music students at the Juilliard School, who are preparing sheet music to help train the model. The goal is to generate music from text or audio prompts, such as creating a guitar track to accompany a song. OpenAI is also considering potential uses for this technology in advertising.
The music industry is wary of these advances. Record labels have already filed lawsuits against Suno and Udio, accusing them of possible copyright violations. While OpenAI CEO Sam Altman has said that rights holders should eventually share in the revenue—a point he raised during the troubled rollout of the Sora app—it's still unclear how this would actually work.
This move marks a return to music AI for OpenAI. Back in 2020, the company introduced "Jukebox," an early experiment in AI-generated music, but hasn't pursued the technology further until now.
Comment
Source: The Information