3 ms·
How can chatGPT be used to do this? Did they ask it to just produce training data?
by bawana 3y ago
How can chatGPT be used to do this? Did they ask it to just produce training data?
- Jensson 3y agoYou can use it to produce the fine tuning examples. LLMs are very good at learning styles and answer flows, so you ask ChatGPT a million standard questions and problems and then record the outputs, that data will be very similar to the finetuning data used by OpenAI. Then put that data into your own LLMs fine tuning step and voila, you have an LLM that behaves very similarly to ChatGPT! OpenAI knows this which is why they put that TOS there, they spent a lot of money to create that finetuning data so they don't want to give all of that away for cheap.
- selfhoster11 3y agoThey likely didn't use (just) ChatGPT, but the GPT-3.5/GPT-4 API. Both can be used to create training data quite successfully. This technique has been used in the past to create synthetic post-training (fine tuning) datasets like Orca, Samantha, and so on.