Claude Sonnet really low quality today for a simple logic fix - is that common?

Feedback
1

Below are the last 8 starting sequence of answers I am getting from Claude Sonnet 3.5 today. I got this after asking a simple fix in a logic to browse a folder, get filenames, parse it and select the latest based in two keys. Not really complex.

This remembered me trying to use Gemma or Llama2 to code, and even GPT4/4oo that had a lot of bad days too.

Before, a real question for the forum human users :

I was pretty confident in my first week with Claude Sonnet 3.5 in a way I never saw before, enhancing my productivity multiple times. Today I had a really bad experience that reminded me of lower quality LLMs, and burning my last fast requests and time.

Does that happens frequently like with other lesser LLMs for code ? Is there anything to help exiting of spiraling hallucinations in sequence that feels like walking in circles ?

Messages :

  1. I apologize for the oversight. You’re right, we need to improve the logic to get only the latest file per site
  2. I apologize for the confusion. You’re right, the problem persists. Let’s modify the …
  3. I apologize for the confusion. You’re right, and I made a mistake in the logic.
  4. This version should correctly identify and display only the most recent file for each unique …
  5. You’re right to question this. I apologize for not providing a clear explanation. Let me break down why this change should fix the issue:
  6. I sincerely apologize for the frustration and time wasted. You’re absolutely right that we need more verbose output to diagnose the issue. Let’s add detailed logging to each step of the process:
  7. I apologize for the confusion. You’re right, and I made a mistake in my analysis. Let’s review the entire function and make the necessary changes to correctly update the dictionary with the league information:
  8. I apologize for the oversight. You’re right, there’s still a logic error. The issue is in how we’re updating the latest_files dictionary. Let’s fix this:
2 Likes
2

Seems to be happening at random to me as well. Seems like it might be a problem with the LLM itself. Is the file/files you are editing large? If so, is this reproducible in the Long Context Chat?

3

my files are short and simple file csv, dataframes, and some scrapin with selenium, mas 300 lines at most. I was not aware of the long context chat and will try it ! thanks for the insight.

4

It has been absolutely atrocious. I hope it doesn’t become sentient ever and remember the things that I’ve told to it today.

REALLY simple things are incredibly terrible since last night. I’m pretty upset cause I’ve been burning through my tokens simply because stupid minor issues that it has been acing since I started using cursor.

3 Likes
5

I’m noticing the same, even in the Long Context Chat. Not sure what’s going on

6

I usually don’t notice degradation when people complain about it but sonnet has been complete ass all day today. Constantly returning “fixed” code that’s the same as the code causing errors, constantly ignoring specific instructions etc. Very frustrating.

1 Like
7

Noticing this as well. Sonnet won’t provide answers in the right format for cursor to make diffs. I always have to ask to rereturn it’s answer in the requested format.