AI Models and Tools Update: ChatGPT’s Tasks, Adobe’s Bulk Edit

OpenAI’s ChatGPT is becoming the little engine that could. Its latest capability: Letting users schedule tasks for a later, or recurring, date. Scheduled tasks is rolling out in beta for users of ChatGPT Plus, Pro and Teams. Enterprise users will get access soon; free users will also eventually get it. OpenAI did not give out specific dates.

With tasks, you can tell ChatGPT to do things like summarize the week’s top national news at 5 p.m. on Fridays, or set up a daily 15-minute workout that focuses on lifting weights. It also can give you recurring reminders, like take out the trash on Sundays at 6 p.m. ET. But make sure you turn on notifications under settings. Users can edit or cancel the task.

Scheduled Tasks is available on the ChatGPT website, iOS, Android and macOS. Integration into the Windows app is coming in the first quarter. For now, the list of tasks you create is only visible on the ChatGPT website. Tasks use OpenAI’s GPT-4o model; users are limited to 10 tasks at a time. In addition, each plan’s messaging limits also apply; Plus members can send up to 80 messages every three hours on GPT-4o, for example.

Stability AI, one of the co-developers of image generation model Stable Diffusion, and Nvidia announced SPAR3D, which can turn a 2D image into an editable 3D object in less than a second. This model can transform 3D prototyping for game developers, product designers and metaverse builders.

Researchers used a two-stage method for the model: the point sampling stage and the meshing stage. The first stage uses diffusion models to generate sparse point clouds, while the second stage transforms these point clouds into highly detailed meshes. That means an uploaded 2D image of an object is changed into little points that follow the shape of the object. Then, it is overlaid with a generated mesh to make it into a 3D object. What’s unique about this process is that users can edit the image.

Adobe introduced a new generative AI tool that can edit large batches of images with one click, according to The Verge. Called Firefly Bulk Create, the app lets users quickly resize up to 10,000 images — or do things like replace the background of images en masse. This helps creative teams apply a similar look to digital assets to maintain brand consistency. The tool is available to customers of the Firefly premium services.

Microsoft researchers introduced rStar-Math, a reasoning method that, when applied to small language models, can match or even surpass the math reasoning capabilities of bigger and more powerful models like OpenAI’s o1, without having to copy knowledge from these larger models.

In a paper, researchers said rStar-Math was able to achieve this performance by exercising “deep thinking” through Monte Carlo Tree Search, where an algorithm builds a “tree” of possible moves and outcomes, simulates plays to see which ones lead to good results, and learns from it.

For businesses, it means they can rely on smaller, more cost-effective AI models to solve complex problems, such as optimizing logistics and predicting financial outcomes. Companies can lower what they spend on AI compute while achieving similar or better results from smaller models. Since rStar-Math can also generate high-quality training data, businesses can use it to develop AI tools to tackle specific tasks like detecting financial fraud.

The post AI Models and Tools Update: ChatGPT’s Tasks, Adobe’s Bulk Edit appeared first on PYMNTS.com.