Hey folks! Imagine this. In a few years, snagging a doctorate or even a Nobel Prize might just get a bit easier. Crazy, right? But why do we think so? Well, it’s all thanks to OpenAI and their latest project, codenamed .
This idea, previously known as Q*, came about because the AI (Artificial Intelligence) models we have today can’t quite tackle those tricky science and maths questions. Their reasoning skills are just not up to par. Plus, they tend to hallucinate – and no, not in a psychedelic way! You see, while these large language models are great at summarising texts, crafting beautiful poems and explaining tough concepts faster than we can, they struggle with common sense problems that seem obvious to us.
Think, recognising logical fallacies or playing tic-tac-toe. When faced with these, they often spit out completely made-up information. This happens because AI models are trained on existing data, and if this data is incomplete or biassed, they pick up on that and learn the wrong patterns.
This is why they sometimes make incorrect predictions or, as we like to say, hallucinate. Enter Strawberry. The goal here is to fix these issues.
Imagine an AI that, after its initial training, can generate its own training data to fine-tune what it knows. This would help it improve its reasoning skills and understand the world more like we do. With this approach, the AI might stop hallucinating and start solving complex research problems that have stumped us for years.
So, brac.
