Most people won't experience how good the latest LLMs are

Most people won't experience how good the latest LLMs are because they still use them as chatbots.

If you ask Fable back and forth questions it doesn't sound any smarter than Sonnet and usually it's just slower.

If you can, try to give it a bigger problem to solve. Something that would take you a few days. The results are truly impressive across multiple domains I have explored: of course coding, but in researching and defining data structures and complex business processes too. This is not just generating a volume of words, this is really insightful and well argued content to solve a real problem.

Sometimes they say smart people seem dumb in person; I think the new models are like that. They need time and space to flex and work.