[lsy]

The title makes it sound like some new architecture, but this is a blog post where someone likes the results they get sometimes when they fiddle with their input to the LLM to suggest “contemplation”, which apparently makes the LLM generate a large paragraph of highly neurotic text before the answer. There aren’t benchmarks or investigation of the model to see whether it is robust or generalizable so it’s hard to say whether this is useful or not.

[padolsey]

Tbf have you read most of the academic papers on LLMs lately? It's all a lot of boilerplate academic language packaging around "we tried prompting it this way and it did good things". Tho yes I do appreciate some scientific prudence.