this post was submitted on 19 Jul 2026
91 points (97.9% liked)
Hacker News
5193 readers
425 users here now
Posts from the RSS Feed of HackerNews.
The feed sometimes contains ads and posts that have been removed by the mod team at HN.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
"Importantly, our goal is not to estimate the overall accuracy of AI assistants, but to isolate the behavioral effect of making an answer available when that answer is systematically unreliable. Therefore, we use Step 3.5 Flash. Because the advice participants could obtain was usually wrong...."
The following Monday morning the research team returned to find all their systems had been replaced with Temple OS.