Hook

Their other posts in the index, biggest breakout first.
Scientists just built an AI model on the human brain and it's already out performing ChatGPT. The system is called a hierarchical reasoning model or HRM and instead of brute force billions of parameters like GPT-5, it uses just 27 million and 1,000 training samples to think more like us. Inspired by the way different regions of the brain process information at different speeds, HRM doesn't break problems into endless text steps like chain of thought models do. Instead it reasons in burst, fast here slow planning there until it lands on the answer. And the results are wild in one of the hardest benchmarks for testing artificial general intelligence, HRM crushed GPT style models, beating GPT4, antiprock and deep seek on reasoning tasks. It even solved complex pseudo puzzles and maze problems that large language models consistently failed. What's crazy is this thing isn't even a tripod parameter scale. It's just designed differently. Researchers say this brain inspired approach could be a key step towards AI or even beyond. And here's the twist. When benchmark organizers try to reproduce the results, they found the architecture itself wasn't the only magic. There was also a hidden refinement process driving the breakthrough, whether that's an overlooked hack or the beginning of a paradigm shift. One thing is clear. We may have just seen the first AI that actually thinks more like a human brain. Let me know what you guys think about all this in the comments below and don't forget to follow for your daily AI news.