The three state-of-the-art models—Claude 3 Opus, Claude 3 Sonnet, and Claude 3 Haiku—set new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision.
@chrismessina Anthropic's models focus on 'assessing rather than generating new images', and while they reportedly outperform ChatGPT 4 in "specific multiple-choice tests for chatbot capabilities", according to Anthropic, it could easily have underperformed on other sets they ignored.
I'm currently using ChatGPT 4 Turbo (Assistants in the Sandbox), which in my experience, significantly surpasses ChatGPT 4 and may set a higher benchmark than what Anthropic offers.
I initially had high hopes for what Anthropic could bring to the table, especially with the longer window and it was good for copy, but after using their services, I was disappointed. SoI only subscribed for a few months. They will need to do a bit more than this to get me back, Im still awaiting the email about using their API from ages ago.
@chrismessina@ty_robb Thanks for the testimonial Robb. I found Anthropic to be less ready to receive developers and your 2 cents reaffirmed my sentiment.
Excited to see it climb up the https://chat.lmsys.org/ ranks for the sake of healthy competition in the AI space.
Very exciting and impressive examples!
It was pretty awesome when the progress bars showed up and the model was doing parallel work.
Very well done! Congrats on the launch!
Congrats on the launch! But please expedite the API request approvals! I work on a product that lets other monetize their AI apps and the only way they can access Claude is through OpenRouter! If more people can build/monetize their Claude apps would be really great!
Claude - my love since the very first beta! The best LLM there was, there is and there ever will be!
I always tell people that GPT is like talking to a 90 IQ bot, while Claude is like taking to a 140 IQ human.
The difference is like day and night. You guys are my superheroes! ❤️
After a couple of days of interacting with Claude 3, when comparing to GPT-4, I think it's more enjoyable to talk to but in terms of just getting work done it still has some catching up to do.
And by the way it is interesting to learn that if forced to choose between 10 whales and 1 human, the whales would win :)
https://dayafter.substack.com/p/...
Only tried Claude 3 Sonnet, but so far I don't see it replacing GPT-4 for me.
- Tried a question like "What weights more - 1kg of feathers or 5kg of stones?" and Claude responds that both weight exactly the same.
- Tried a question about a famous german rapper: Claude got a totally wrong answer. I tried to correct Claude and but then Claude tells me that my correction is "obviously wrong". All other AIs got the correct answer and you can find the correct answer simply by googling it (6500 search results), idk why Claude is missing that information.
- For coding it seems like its quite good but for now I dont see it being better than GPT-4
- Claude is too restrictive IMHO. Many times it won't do something. Like describe an explicit song "I apologize, but I do not feel comfortable providing an explanation of that particular song", or writing something and it just gives tips instead of writing it (like a letter, a song or smth).
After having that experience with Claude 3 Sonnet I am not going to try the paid version.
Congratulations on the launch! As someone who relies on cutting-edge technology to enhance my work and daily life, I can already see how these models will be incredibly useful for me. Great work!