Claude 3 is getting Weird (Claude 3 fails vs ChatGPT)

Рет қаралды 1,961

3 ай бұрын

Claude 3 might be the most intelligent LLM out there, but I've found some interesting prompts that make it give strange answers. We'll be comparing Claude 3 Sonnet, Claude 3 Opus, and ChatGPT (GPT-4 + GPT-3.5). Enjoy!
🔗 Links:
Claude 3 - claude.ai
Anthropic - www.anthropic.com
Claude 3 API - www.anthropic.com/api
📧 For business inquiries: business@monice.net
Don't forget to smash that like button, drop a comment, and subscribe for more awesome content like this.

Пікірлер: 14

@moniceai 3 ай бұрын

Thank you for watching! Write down video ideas that you would be interested in watching :)

@virtualalias 3 ай бұрын

"We're gonna putting them in test?" That's funny, I actually though this was a real video for a second there.

@moniceai 3 ай бұрын

What do you mean? I meant that I found prompts to which Claude 3 gives strange answers, and I was going to check if it's really true, or if this happens once in a million, or if it's completely fake.

@wilmarkjohnatty4924 3 ай бұрын

Is this really your voice? Its a very good voice for narration.

@lancemarchetti8673 2 ай бұрын

Interestingly Claude Sonnet and Haiku produce better code than ChatGPT

@poliestotico 3 ай бұрын

Why whyyy is claude 3 US locked :(

@lovechatgpt 3 ай бұрын

This was my gpt4 result straight away"Yes, 450 is 90% of 500. You can calculate this by multiplying 500 by 0.9 (which is 90% expressed as a decimal): \[ 500 \times 0.9 = 450 \]

@moniceai 3 ай бұрын

Amazing. Because I tried this prompt several times and it always first said NO, and then it did the calculations and said YES :)

@JJSeattle 3 ай бұрын

ChatGPT 4 right after this video: Is 450 90% of 500? ChatGPT4, "Nope, 450 isn't 90% of 500. To figure out what 90% of 500 is, you just multiply 500 by 0.9. Let's do the math real quick. Well, butter my biscuit, I stand corrected. 450 is indeed 90% of 500. Looks like the math checks out perfectly!"

@moniceai 3 ай бұрын

hahaha that's funny, I'm very curious about why they respond like that, even GPT-3.5 doesn't make such a mistake xD

@gbrailsford 3 ай бұрын

Come on Monice. The thumbnail for this video is ridiculous.

@moniceai 3 ай бұрын

Why? This thumbnail works well. If you think it's "ridiculous", do you have a suggestion on how to make it better?

@miked7373 3 ай бұрын

Ridiculously good!

@gbrailsford 3 ай бұрын

@@moniceai It's completely misleading about the responses that Claude 3 might provide. It doesn't relate to the content of the video which is about niche incorrect responses by the model, not audacious offensive responses. It's pure click bait.