News for hackers. Stuff that matters.
Re:OpenChat 3.5: 7B model with comparable perf to ChatGPT by fgfm on Thursday November 02, 2023 @04:19PM (#38119506)
The Alignement AI Lab just published OpenChat 3.5, which is outperforming ChatGPT (march version) on most benchmarks apart from MMLU (67.3% vs 64.3%) & BBH-CoT (70.1% vs 63.5%).
Re:OpenChat 3.5: 7B model with comparable perf to ChatGPT by circuit10 on Thursday November 02, 2023 @08:45PM (#38122768)
It seems very good but it thinks that it's ChatGPT:

"Who are you?

I am an AI language model named ChatGPT, developed by OpenAI. I am designed to engage in conversations and provide information on a wide range of topics."

Edit: I can see in the paper they linked it mentions using ShareGPT data