Yes, the previous best LLM* for coding ( I do python mostly) completely wasted my saturday for more the 4 hours with absolutely poor responses. Most of code changes created more problems that took 5 to 10 more prompts to fix or revert making me loose time and point the problem and solution multiple times.
A new annoying behavior of answering partially with a generic text ( in the lines, yes comparing text strings can cause problems like you describe, you should investigate problems in those text strings to solve ) with a worst question
**
Would you like me to explain any part in more detail?
**
asked so many times that I started including in my prompts to not ask it.
WHAT A HUGE TOKEN COST FOR NOTHING.
And finally, the code changes were so low quality that running a local model in my mac is better.
This example is the worst I’v seen :
I asked to add a few similar columns in a existing df built in a test file ( simple stuff ) to compare character unicode representations. With this code in the middle :
rows.append({
'source': source,
'league_name': league,
'character': char,
'unicode_point': ord(char),
'nfc': nfc,
'nfkc': nfkc,
'nfkd': nfkd,
since the first run, I had follow up runs with this repeating errors for the same cause :
KeyError: ‘NFC’
KeyError: ‘nfc’
KeyError: ‘league_name’
KeyError: ‘league’
anyway, I would love to send, share code if that could be used to improve this
LOUSY, ANNOYING and really low quality “day” and consuming my Fast Requests for a lost saturday that I could be with my family.
- ( now I know all of them the quality is variable, which I am quite familiar happening to human beings, and gone are the days, technology were reliable ).