News for hackers. Stuff that matters.

A comment on RightWingGPT – An AI Manifesting the Opposite Political Biases of ChatGPT.

Re:RightWingGPT – An AI Manifesting the Opposite Political B... by EGreg on Sunday February 19, 2023 @04:15PM (#34861936)
"Critically, the computational cost of trialing, training and testing the system was less than 300 USD dollars."

Is this a joke?

What exactly did he do? Training the system how? Does ChatGPT allow APIs, let alone training on some corpus of data?

[ Parent | on HN ]
Re:RightWingGPT – An AI Manifesting the Opposite Political B... by jncraton on Sunday February 19, 2023 @04:20PM (#34861971)
OpenAI provides tools to fine-tune models:

https://platform.openai.com/docs/guides/fine-tuning

The largest Davinci model currently costs $0.03 per 1k of training tokens during fine-tuning, so you could fine-tune a model on 10M tokens for one epoch at a cost of $300.

[ Parent | on HN ]
Re:RightWingGPT – An AI Manifesting the Opposite Political B... by EGreg on Sunday February 19, 2023 @04:44PM (#34862167)
Wow! I can use this for forum moderation ... basically filtering user input through this to create "innocuous" and "more standardized" titles for events etc.
[ Parent | on HN ]