Claude 2 - Anybody got approved yet?

[Nan0]

Just seen that claude 2 has a yummy 100k context window, was wondering if its reasoning abilities and content writing skills are up to par.

 

[lfgPeople]

yep it's very good

 

[SyntaxSurge]

My problem is that it is much more strict than chatgpt

 

[Nan0]

My problem is that it is much more strict than chatgpt

I watched some videos and apparently it's very patronizing, anthropomorphizes itself, and thinks it's the arbiter of moral authority. Plus in the meantime by trying to make it safe they've learned it to lie. I don't need that from a chatbot.

But for the purpose of spinning articles it might seem like a reasonable option if you're churning a lot of data.

 

[rb2023]

I've never had an issue with Claude doing any of the things that some of the people on YouTube claim that it does. Personally, the best use of Claude 2 at present seems to be writing blog posts based upon a lot of background knowledge and avoiding AI detection (Originality doesn't know how to detect Claude yet). It is quite far behind GPT-4 in a lot of the other areas. They're also pretty slow at giving API console access and production API keys.

 

[DeFiNFT1]

not a fan myself

 

[rb2023]

In fact, you can think of Claude and gpt4 as the difference between liberal arts and science. Aside from the 100K length, Claude is actually stronger in the liberal arts, it's better written. However, in logic, i.e. science, it's obviously not as good as gpt4. I have a spare Claude api account, so if you need it, you can bring your price to me privately.

I don't think Claude is better written at all. It's not as descriptive, it's not as capable of writing in different styles and it isn't as creative. It can beat AI detectors better, but that's because the detectors aren't trained to detect Claude yet, not because it's better written or more "human" (as some have suggested). It also isn't drastically worse than GPT-4 at logic, it is worse, but not by much. It may seem it from the Web app (which is set at a very high temperature) but if you reduce the temperature with the API it is a lot stronger at those kind of tasks.

The main benefit is the context window, which allows it to do some impressive things in longer conversations or with more background knowledge. But the lack of a system prompt or some other way of giving a stronger weighting to certain parts of the text really limits the power of the what you can do with the context window.

Claude 2 is the nearest competitor to GPT-4 and it's not a million miles away in all aspects, but at the moment the only benefit is the context window and ability to beat AI detectors. But I don't think it will have that advantage for long. And I do think they will catch up in future versions.

 

[ninicky111]

I don't think Claude is better written at all. It's not as descriptive, it's not as capable of writing in different styles and it isn't as creative. It can beat AI detectors better, but that's because the detectors aren't trained to detect Claude yet, not because it's better written or more "human" (as some have suggested). It also isn't drastically worse than GPT-4 at logic, it is worse, but not by much. It may seem it from the Web app (which is set at a very high temperature) but if you reduce the temperature with the API it is a lot stronger at those kind of tasks.

The main benefit is the context window, which allows it to do some impressive things in longer conversations or with more background knowledge. But the lack of a system prompt or some other way of giving a stronger weighting to certain parts of the text really limits the power of the what you can do with the context window.

Claude 2 is the nearest competitor to GPT-4 and it's not a million miles away in all aspects, but at the moment the only benefit is the context window and ability to beat AI detectors. But I don't think it will have that advantage for long. And I do think they will catch up in future versions.

It might have something to do with my native language, I'm a Chinese, and maybe Claude focuses more on Chinese corpus in his training than gpt4. In some ways, I do think it has more subtle emotions in its delivery. Of course I do agree with you that in every way it is indeed inferior to gpt4, although gpt4's brain is slipping right now too.

 

[medialounge]

Claude 2's ability to beat AI detectors is not a sign that it is better written or more human.

 

[frozen_yoda]

I have access directly through Anthropic. Paid API requires a partnership, so unless if you're willing to spend a lot of money that's not really an option.

In fact, you can think of Claude and gpt4 as the difference between liberal arts and science. Aside from the 100K length, Claude is actually stronger in the liberal arts, it's better written. However, in logic, i.e. science, it's obviously not as good as gpt4. I have a spare Claude api account, so if you need it, you can bring your price to me privately.

You can't sell outside of the marketplace. Not to mention you're trying to resell something that is easily accessible.

----

Everyone: Sign up to use Anthropic in AWS - Amazon Bedrock. Depending on region, it's immediately approved.

GCP also has it in their model garden.

There are lots of gen AI models in both places. GCP has a lot more. I use both, but we have contracts and NDAs in place.

Azure has a lot as well and we have contracts with them too, but they do not offer Anthropic.

All of these have API access and are great for testing. Huggingface models are easy to deploy on GCP and Azure as well.

 

[flackz]

Still waiting for a approval for a Claude API key...

 

[frozen_yoda]

Still waiting for a approval for a Claude API key...

Use AWS or GCP and get access immediately.. You won't get Claude API access directly through Anthropic without a contract to spends tens of thousands of dollars.

 

[Binkmaster]

Plus in the meantime by trying to make it safe they've learned it to lie. I don't need that from a chatbot.

They taught it to lie! lol