OpenAI’s ‘Strawberry’ o1 AI Can Think Like a Human—But Why Is It Named After a Fruit?

news7g2 hours ago

0 2 minutes read

OpenAI's 'Strawberry' o1 AI Can Think Like a Human—But Why Is It Named After a Fruit?

When you think about it artificial intelligence models, you can envision tools that make work more convenient, enhance the functionality of smartphones, and generate text or images in seconds. However, in the race to achieve AGI (artificial general intelligence) and improve reasoning abilities, companies like OpenAI have shifted their focus to improving the model’s cognitive skills. OpenAI, the creators of ChatGPT, have now announced a new model called o1, internally known as ‘Strawberry.’ Along with that, they introduced another model, the o1 mini—designed specifically for cryptographic tasks and priced 80% lower than the o1. But what makes this new model different from others, like OpenAI’s GPT-4o and GPT-4, and why is it called ‘Strawberry’? Read on to find out more.

Why is OpenAI o1 internally called ‘Strawberry’?

If you open ChatGPT If you use GPT-4o and ask it how many times ‘r’ appears in ‘Strawberry’, it might give you an incorrect answer and answer twice, which is clearly wrong because the correct answer is three. It’s this irony that reminds us that AI may not be as smart as we think it is after all—there’s very little thought put into answering your questions. This may be exactly why OpenAI uses the name ‘Strawberry’ internally—a project that develops an AI model that spends more time thinking, despite requiring more computing power, time, and financial resources. In fact, Sam Altman, CEO of OpenAI, posted a playful Easter egg ahead of the official reveal earlier this month when he shared an image of a strawberry plant on X (formerly Twitter).

OpenAI o1 Model aka ‘Strawberry’: What’s New?

So far, with models like GPT-4 and GPT-4o, OpenAI’s focus was on solving problems quickly and providing quick responses to a variety of prompts. With o1, however, OpenAI claims to have created a model that spends more time thinking before responding. Why? It needs to reason through complex problems and even come up with solutions in a variety of domains, such as science, mathematics, and coding. For example, it can handle tasks like annotating cell sequencing data or generating the complex mathematical formulas needed for quantum optics, etc.

OpenAI claims that the model is similar to how a person thinks. Through training, it refines its errors, develops its reasoning ability, and most importantly, recognizes the mistakes it makes. “In our experiments, the next update to the model performed similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology. We also found that it excels at math and coding,” OpenAI said.

The model is still in its early stages, though, which is why OpenAI decided not to let it access the web to fetch information or upload images and videos. For most tasks, OpenAI says, models like GPT-4o will still be more capable, but if you want a model that thinks more about every key query, the o1 (‘Strawberry’) is the way to go.