My actual problem isn't with quality, it's speed. I either have to put in very specific prompts and wait for a good model to think, or iterate many times with a lesser one with incremental prompt refinement. Neither of these are effective uses for me. Programmers may be a minority but others who have specific uses with high bar demands must also exist. As far as I can tell from reading HN, programmers are a high value target so would be odd to not be well represented.

With the number of bad studies, it might be a better default to consider all posts not well-founded until repeat independent studies appear.

The small sample size probably is the major factor, that along with its subjective summarization.