In the world of technology, scientists have made a big step forward in teaching computers how to think better. 

A team from a company called Google DeepMind and a university named the University of Southern California have built a new tool that helps computers, specifically those that use many words, to solve challenging problems more like humans. Let’s dive into what they did and why it’s so exciting!

---

• Google DeepMind and USC Create “SELF-DISCOVER”: A game-changing framework that teaches AI to think and improve problem-solving by up to 32%.
• AI Mimics Human Reasoning: The new approach allows computers to autonomously develop and apply step-by-step reasoning akin to human thought processes.
• Advances Toward General AI: This innovation marks a significant step towards developing AI that can understand and reason through complex problems like humans.

---

AI Takes a Leap in Problem-Solving with New Google DeepMind Framework

Intelligent people from Google DeepMind and the University of Southern California are making waves in the tech world. 

They’ve created a clever tool to help large language models (LLMs) – computers that work with lots of words – think and solve problems much better. This tool is like a new set of instructions that could change how we use artificial intelligence (AI).

Key Developments:

• SELF-DISCOVER Framework: This is the name of the new tool. It’s been talked about in two places: arXiV and Hugging Face. It’s a big deal because it could change how computers solve tricky problems by thinking smarter.
• Enhanced Performance: When they tried this tool with famous computer models like GPT-4 and PaLM 2, they found up to 32% better at solving problems than the old ways.
• Autonomous Reasoning: Now, computers can figure out how to tackle challenging problems by following steps, just like a human would.
• Methodology: There are two steps in this new method. First, the computer makes up a plan for how to think about the problem. Then, it uses that plan to come up with an answer.
• Test Results: They tested this new tool with different brain teasers and puzzles. The results were impressive: it got 81%, 85%, and 73% of the problems right, which is way better than before.
• Broader Implications: This isn’t just about computers being better at solving puzzles. We’re moving toward computers that can think like people, a significant technological dream.

The “SELF-DISCOVER” tool lets computers find and use little bits of logic and thinking steps to work through problems. It’s like teaching a computer to solve a puzzle as a person would, starting from scratch and building the steps to find the answer.

They tested this new method with many challenges, and it did well! It beat the old ways of doing things, especially in games and tests that needed innovative thinking and solving math problems. 

Thanks to the “SELF-DISCOVER” tool, computers can now handle more complicated tasks. This is a big step forward, making computers more thoughtful and valuable.

As computers improve, discoveries like the “SELF-DISCOVER” tool are super important. They help us move forward, making new things possible and showing us a sneak peek of what computers might do in the future.