3 comments

  • heaney-5551 minute ago
    &gt;We study a frontier LLM in its purest inference regime - a single generation pass over a prompt containing the full training and test data, with no tools, no agentic scaffolding, and no fine-tuning<p><i>Sigh</i>. So this is somewhat interesting niche academic research but utterly irrelevant to real-world use cases.
  • ismailmaj38 minutes ago
    Unsure if it&#x27;s LLMs that fail at tabular data or its just that tree boosting are spectacular at that task.
  • cyanregiment1 hour ago
    Just have 2 LLMs debate whether tabs or spaces are the superior choice