GPT-4 Turbo as judge and polar plot script

[lionelchg]

Why are these changes needed?

Since GPT-4-Turbo is the new standard that tops GPT-4, it should be added to the list of available judges. Doing so changes the actual scores of LLMs which are lowered:

Model	GPT-4 Judge	GPT-4-Turbo Judge
GPT-3.5-Turbo	7.944	6.750
GPT-4	8.991	7.743
GPT-4-Turbo	9.259	9.412

The corresponding polar plots for GPT-4 and GPT-4-Turbo are as follows:

• GPT-4 Judge:
  

• GPT-4-Turbo Judge:
  

A script polar_plot.py has also been added with a json configuration file to easily produce the polar plots.

Checks

• I've run format.sh to lint the changes in this PR.
• I've included any doc changes needed.
• I've made sure the relevant tests are passing (if applicable).

[infwinston]

Thanks @lionelchg! This is very useful information. We'll explore whether gpt-4-turbo as judge works better.

For this PR, we may want to keep llm-judge main directory minimal. Could you create a misc directory and move the plotting script and all the configs under it?

[lionelchg]

I have put all the config files and scripts inside a misc/ directory and updated the README.md accordingly 😄