ChatGPT-3.5 Turbo is leading in the LLM Colosseum benchmark for Street Fighter III, outperforming iterations of ChatGPT-4 with an Elo rating of 1,776.11.
The success of ChatGPT-3.5 Turbo in gaming signifies the importance of a balance between speed and intelligence in LLMs, as noted by Nicolas Oulianov, one of the developers.
Oulianov highlights that the disparity between ChatGPT-3.5 Turbo and GPT-4 reflects the prioritization of certain features in the latest LLMs, emphasizing the need for custom evaluations for specific use cases.
A different experiment by Amazon Web Services developer Banjo Obayomi showcased varied LLM behaviors in the Street Fighter III benchmark, with model Claude standing out as the top performer scoring first to fourth place.
[
add
]
[
|
|
...
]