OpenAI has unveiled GPT-5, its new flagship model that will now power ChatGPT. GPT-5 is positioned as the first “unified” model that combines the reasoning capabilities of the O-series models with high-speed regular models. Instead of asking you to select the right settings, GPT-5 itself determines how to offer the best response: by quickly answering or spending additional time “thinking” about the answers. (But as you know, answers after thinking are just as hallucinatory as without them, but now you can’t turn off reasoning - you can only constantly click Quick Reply). In addition, the new model (unlike GPT-4) is reportedly optimized for creating not chatbots, but agents capable of performing a wide range of tasks on behalf of the user.

Atlassian has introduced an update to its project management service Trello and it was immediately met by users in a sharply negative way. The update introduced a variety of changes to the service interface, for example, it changed the arrangement of all the main interface elements on the page. The fonts and functions of the buttons have also changed - especially in the task card. Many users are sure that the new Trello interface was implemented specially to make it as inconvenient as possible and it's a secret attempt by Atlassian to force them to switch to Jira, which provides similar functionality, but is more expensive. Both services have free plans, but if you consider the paid plans, Jira costs at least $7.53 per user, while Trello costs $5. As a reminder, Atlassian acquired Trello back in 2017.

SAP has acquired recruiting management software developer SmartRecruiters. In its press release, SAP says that SmartRecruiters' "powerful, easy-to-use interfaces and workflow engine" will complement SAP's existing HR tools (apparently referring to the SAP SuccessFactors). The system is also known for its AI assistant called Winston, which automates most routine tasks and generates useful analytics. With the acquisition, "SAP customers will be able to manage the entire candidate lifecycle - from sourcing and interviewing to onboarding and training - all in a single system, making things easier for recruiters. SmartRecruiters was founded in 2010 and has raised $225M in funding over 15 years.

Adobe has added new image editing and enhancement features to Photoshop based on the Firefly AI model. The Generative Upscale feature allows to improve the image resolution up to 8 megapixels. This is useful for restoring details in old photos, scaling images for printing or adapting assets for different platforms. The Harmonize feature allows you to place a new object in a scene, remove its background, adjust the colors, lighting and shadows to match the object to the scene. The idea is to make the new object, such as a chair in a room, fit harmoniously into the environment. Adobe has also introduced an improved Object Removal tool. Previously, when using generative fill to remove an object, the tool took into account the entire scene to fill the gap left by the removed object, which sometimes led to unwanted artifacts in the image. The company has reworked the algorithm so that it focuses only on removing the object locally.

Cybersecurity giant Palo Alto Networks is buying identity management and protection company CyberArk. The deal is one of the biggest in the history of the IT market - $25 billion. And, naturally, this is the largest acquisition for Palo Alto Networks, which previously bought Dig Security ($400 million in October 2023), Talon Cyber Security ($625 million in November 2023) and Bridgecrew ($156 million in February 2021), among others. Palo Alto CEO Nikesh Arora says that with this acquisition, the company enters the identity security market and gets a platform that can solve the looming problem of AI agents that learn to authenticate instead of people.

Microsoft has not stayed aside from the AI browser craze and has launched a new feature called Copilot Mode in its Edge browser, which allows users to work with web pages using AI. The idea is that AI can become an assistant that understands what the user is looking for, predicts their actions and performs actions on their behalf. Copilot appears in a modal window on top of the page and can be asked in the context of the content of the current page. For example, the user can ask the AI assistant if the recipe they are viewing can be made vegan. In addition, Copilot can perform various tasks on behalf of the user, such as booking an appointment, creating shopping lists and (of course) writing content. Copilot supports voice input, so you can command via voice. The company emphasizes that Copilot will be able to access the browser content only with the user's consent.

MIT developers have created and published on GitHub an open-source neural network GenCAD for generating 3D models from photos. All you need to do is upload an image that you want to turn into 3D model and it will handle it. In fact, GenCAD automates the work of design engineers, allowing them not to waste time creating simple models in AutoCAD and other CAD systems. GenCAD is not the first service of its kind, but it has several fundamental differences from its main competitors. The most important of them is that most CAD generators are limited to creating only a polygonal mesh of the model, while GenCAD goes all the way to building a 3D model.

OpenAI has introduced ChatGPT Agent, an AI agent that can navigate the web and perform tasks on your behalf. It can search the web and collect information, scroll pages, click buttons and submit forms. For example, it can go to your GMail and answer your question about who sent you new emails. In fact, it works similarly to the browser assistant Perplexity Comet. Anthropic also has a similar tool called "use computer" that can control the cursor and write code. OpenAI and Anthropic are expected to introduce browsers with built-in agents in the near future.