Kobold84 wrote:
@Behrauder
Share your opinion about gpt4.5 and claude3.7.
GPT-4.5 is currently only available for pro users (I'm on Plus). But based on other people's tests, it seems to be
the most human-like model yet and also
the most creative one. However, that's about it. It’s not a reasoning model, so it won’t excel in benchmarks.
It’s not as impressive as expected (and it's extremely expensive), but its reduced confirmation bias is a great improvement. This model will be
very useful for training future reasoning models, such as a potential o4 (though I doubt that will be the name).
As for Claude 3.7, Anthropic doesn’t seem to care about free users, and arguably, it doesn’t care much about anyone who isn’t using the API. Claude is clearly an LLM designed for developers rather than general use.
For frontend development, it is far better than any other LLM, and I don’t see that changing anytime soon.
The problem is that if you request even a simple code (maybe what I consider simple isn't exactly simple), you'll likely hit your message limit after just two uses (and from what I understand, you'll have to start a new chat...). Because of this, Claude isn't really usable unless you’re accessing it via the API.
I'm excited to see what DeepSeek R2 will be able to do.