I am writing an essay on the upcoming 2024 US presidential elections. And I copy and paste a news report from BBC (https://www.bbc.com/news/articles/cx28mpzlz47o) into Grammarly to see how it's going to do the grammarly check and correct the published articles. Grammarly gives two suggestions: 1. Change "vice-president" to "vice president". 2. Correct "seized on the furore" to "seized on the furor", changing the word "furore" to "furor". Since I am not familiar with the phrase "seized on the furor", I asked Chatgpt what does it mean and got the answers that "The phrase "seized on the furore" means that someone, often a public figure or media outlet, took advantage of or capitalized on a situation of intense public interest or controversy (the furore). In a news context, it typically suggests that they are using the excitement or outrage surrounding an issue to draw attention to their perspective or agenda." So the original news report text are correct and the rare phrases "seized on the furor" should not be corrected.

I used the abstract text from Transformer paper "Attention is all you need" and Diffussion model paper "Denoising Diffusion Probabilistic Models" to test the ability of Grammarly. Grammarly does provides some advice on the text, such as adding hyphen to "high quality" to "high-quality" and CIFAR 10. But it seems a little bit too obvious and it doesn't give a lot of advices on the expression changes. I will give a moderate rating to the performance. Maybe the premium version will be better?

Price is $3 per million input tokens $15 per million output tokens. Still a little bit expensive in performing complex tasks.

Claude Opus 4 claims that Claude Sonnet 4 achieves strong performance across SWE-bench for coding, TAU-bench for agentic tool use, and more across traditional and agentic benchmarks. It's astonishing what's the performance compared to OpenAI O4 and other models?

Claude 4 is the most exciting model reason I am expecting in 2025 since OpenAI stop release new capable models. Its coding and AI agents capability is the most desirable features of future workflows and AI automation. Hopefully the API price will not increase too much.

Claude gives the correct answer to the math problem of finding the differentiation of a function. The result is very similar to the ones of Gemini for the same question "differentiation of function f(x) = e^x + log(x) + sin(x)?". And it also give the restriction of x>0 because there is a function log(x) in the original function. So it's pretty helpful.

Claude answered my math question of solving a quadratic equation. It uses the quadratic formula and set the coefficient a=1, b= 5, c= 6 to the quadratic formula and gets the exact answer. The proof and the step by step solution is exactly correct and very much helpful!

Claude does an amazing job writing code for the trap water Leetcode. It also provides explanation of the algorithm complexity.

I've tested the AI model implementation ability of multiple AI Code Generators. And for this prompt "Self Attention Layer in Transformer", Claude provides the best overall response. The pytorch implementation of Transformer model's self attention layer is correct. And the detailed explanation is really helpful. For example, it introduce the breakdown of the python code, including the "__init__" method, the forward method, the input tensor, the attention energy function, etc. And I will also attach the detailed text response here: """ Claude Sonnet 3.5 Here's a breakdown of how this code works: The __init__ method initializes the linear layers for the value, key, and query projections, as well as the final linear layer that combines the multi-head outputs. The forward method takes in the values, keys, query, and an optional mask tensor. The input tensors are reshaped to have a separate dimension for the number of heads, so that we can perform multi-head attention in parallel. The value, key, and query tensors are projected using the respective linear layers. The attention energy is computed using the einsum operation, which performs the scaled dot-product attention operation across all heads in parallel. If a mask tensor is provided, the attention energies corresponding to the masked positions are set to a very large negative value, so that they get zeroed out after applying the softmax. The attention probabilities are computed by applying softmax to the attention energies. The values are weighted by the attention probabilities using another einsum operation, and the results are concatenated across the head dimension. Finally, the concatenated tensor is projected using the final linear layer to obtain the output of the self-attention layer. You can use this SelfAttention module as a part of your Transformer architecture by applying it to the input embeddings and optionally using the output as the input to the feed-forward network. """

Claude gives me several reason of hiring personal injury lawyers. But it doesn't give more information on the reasons of not hiring a lawyer, not it's tailed to my specific question "victims of car or truck accident". So I will only gives it an average rating. Not very helpful, and I still need to search for more information after asking Claude this question.