7. Resolving the three issues with ClaudeAI

7.1. Introduction

In [1], the URL from the IA ClaudeAI [https://claude.ai/chat] product from the company Anthropic [https://www.anthropic.com/];
In [2], your chat history. The free sessions of ClaudeAI are very limited. I purchased a one-month paid subscription to conduct the following tests;
In [3], your question;
In [4], to attach files to your question;
In [5], to submit your question;

7.2. Problem 1

The question:

ClaudeAI answers correctly.

7.3. Problem 2

The question:

I have attached two files to my question:

The PDF generated by ChatGPT [Le problème selon ChatGPT.pdf];
My instructions in the text file [InstructionsAvecPDF4.txt]. This is the one that enforces the 25 unit tests proposed by ChatGPT;

The first response is incorrect. Here are the execution logs:

The answer still contains an error, but it is minor. ClaudeAI fails a test by 2 euros, whereas the required precision for the tests is 1 euro. In fact, at the time, both Gemini and ChatGPT had failed this test for the same reason. It is likely that the 1-euro tolerance constraint is too strict due to rounding issues for which the official rules are unknown.

In any case, after two additional rounds of back-and-forth, ClaudeAI provides the correct solution.

7.4. Problem 3

The question:

In [2], we attached a text file already used with the previous IA. It forces IA to search for its information on the internet and again imposes the 25 unit tests.

The first response has many errors. We send the logs to IA:

Still errors:

Still not working. We encourage it:

Still not working:

No luck:

Still a lot of failures. We’ll assume that ClaudeAI was unable to solve problem 3 within a reasonable time.