Yeah the biggest issue is that if the model doesn’t see the answer in the text you show it it’s likely going to guess (hallucinate) and it can be really hard to talk it out such hallucinations. Even with the best grounding prompts there are just some hallucinations you can’t talk it out of.
stevenic
41
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Open AI embedding with 3072 dimensions | 24 | 6503 | July 4, 2024 | |
| Scaling RAG chatbot system to millions of documents | 18 | 7605 | February 28, 2024 | |
| The length of the embedding contents | 47 | 36283 | November 2, 2023 | |
| Train (fine-tune) a model with text from books or articles | 62 | 29572 | November 30, 2023 | |
| Should I modify user queries before semantic search? | 22 | 4516 | June 28, 2024 |