OpenAI is working on a new model named Strawberry. This model might be a big step up from their current AI, GPT-4. Strawberry has shown it can solve tough math problems and even complicated word puzzles.
The launch date for Strawberry is still not set, but it could be soon. Many think it will be ready by November. Strawberry is designed to solve problems it has never seen before. It can also help with programming and marketing strategies. This is a big deal because most AI models today cannot do this well.
Strawberry is different because it uses something called "test time compute." This means it takes extra time to think about a question before answering. It can also answer more subjective questions, like giving advice on product marketing. This makes it a "Reasoner" model. It thinks harder and better than other models.
OpenAI is also working on another model called Orion. Orion is meant to be the next big thing after GPT-4. It will use data generated by Strawberry for training. This synthetic data helps the new model learn without needing tons of real-world data.
OpenAI wants to stay ahead of its rivals, like Google and Anthropic. These companies are also working on AI that can solve complex math problems. Google's DeepMind, for example, is focused on solving math and geometry issues. Anthropic's latest model, Claude 3.5 Sonic, is great at interpreting charts and graphs.
One of the main issues with AI models today is reliability. For example, if an AI makes a mistake early in a task, it can mess up the whole process. This is why OpenAI is working hard to make Strawberry reliable. A reliable model is crucial for tasks that need high accuracy, like coding or customer service.
OpenAI might release a simpler version of Strawberry for ChatGPT. This distilled version will be easier and cheaper to run. It will also help OpenAI keep its top spot in the AI world.
In short, OpenAI is making big moves with Strawberry and Orion. These models are designed to solve problems better and think more deeply than current AI. This could change how we use AI in the future, making it more reliable and capable.