Hacker News Reader

For me the point when things started to went south is when they swiched from gpt4 to phind-70b as driver behind their answers.

I would rather have exected they make a deal with OpenAI or Anthropic, ... and use one of their models at discount while improving the search functionality.

View on HN

I used Phind quite a bit about a year ago and thought they were actually very good. In comparison to Perplexity, for example, I thought results were simply much better. I guess at some point the approach of finetuning their own models could not scale, and using third-party APIs would just be prohibitive as well.

Congratulations to the team though, for what it's worth I really liked the product.

Reply on news.ycombinator.com