Hacker News Reader

Comment by jerpint | Hacker News Reader

I’ve been noticing this too. The other day, cursor’s version of Claude sonnet (3.7) added a

with open(file) as f:

  pd.read_csv(f)

This was a mistake not worthy of even gpt3… I’ve also noticed I get overall better suggestions from Claude desktop app.

I wonder why

View on HN

I cannot find it anymore, but there is a youtube video of someone making / training an llm who showed a lower precision switch switching his small model from bad to lobotomised and he even mentions he thinks it is what Anthropic does when the servers are overloaded. I notice it, but have no proof. There seems to be a large opportunity for screwing people over without consequences though, especially when using these API's at scale.

Reply on news.ycombinator.com