Sign In

Eleanor Berger

Highlights

Aug 29 · linkedin

New "style control" in the LMSys arena is super interesting. It strips away the effects of output style (for example: formatting with markdown) on human preference to reveal a ranking that more faithfully represents model intelligence. As expected, smaller models drop significantly, and some models that are famous for having worse "style" rise closer to the top.

I'd love to see benchmarks and arena matches focused on tool calling and structured output. That's where the power of LLMs really shines.

Aug 29 · linkedin

Related Questions

What is the person's full name?
Is a middle name provided?
Are any affiliations or organizations listed?
Are contact details or locations provided?
Eleanor Berger
Add to my network