Because all the important ones have been true since 2022 and will be true forever, because they have nothing to do with LLM capabilities and instead are focused on LLM nature.
Because they haven't changed yet (despite AI bros claiming that they have every 6 months), and therefore are unlikely to change as we move into the future. It's not impossible, of course. But when the "this new model is totally better" claims keep turning out to be false, that doesn't give confidence that the future is going to be different.
But when the "this new model is totally better" claims keep turning out to be false
Current models are in fact much better than the top models from a year ago. I have no idea how anyone could claim otherwise without a huge amount of motivated reasoning.
