u

posts/llama_index_rag.ipynb

@@ -382,15 +382,15 @@
     "\n",
     "1. Question is about the materials used to construct the Rhodes statue.\n",
     "\n",
-    "The standard LLM claims that wood was used in the construction of the Rhodes statue, but there are no records in the Wikipedia about that. The RAG answer is factual correct, it mentions bronce. The Wikipedia article also mentions other materials, but either the LLM missed those.\n",
+    "The standard LLM claims that wood was used in the construction of the Rhodes statue, but there are no records in the Wikipedia about that. The RAG answer is factual correct, it mentions bronce. The Wikipedia article also mentions other materials, but either the LLM seemed to miss those. In earlier tries I got the system to list all four materials mentioned in the Wikipedia article by varying the question and perhaps simply because of the random seed used. \n",
     "\n",
     "2. Question is about the pose of the Rhodes statue.\n",
     "\n",
     "The standard LLM gives a lot of hallucinated detail. We don't know much about the pose, and the short RAG answer summarises that.\n",
     "\n",
     "3. Question is about whether the statue was clothed or naked.\n",
     "\n",
-    "The standard LLM says it was clothed, probably because a lot of the antique statues were clothed, but the RAG answer is correct, the statue was naked as far as we know.\n",
+    "The standard LLM says it was clothed, probably because a lot of the antique statues were clothed, but the RAG answer is correct, the statue was probably naked according to the Wikipedia. We can see here that RAG indeed overrides what the LLM would normally say.\n",
     "\n",
     "4. Question is about the city in which the Hanging Gardens were supposed to be located.\n",
     "\n",
@@ -411,6 +411,7 @@
     "- Smarter text segmentation based on semantic similarity or logical document structure.\n",
     "- Postprocessing the retrieved documents, by letting a LLM rerank them according to their relevance for the query.\n",
     "- Asking the LLM to critique its answer, and then to refine it based on the critique.\n",
+    "- Generate multiple responses from the LLM and then let the LLM summarize them. \n",
     "- ... and many other ways, it is an open and active field.\n",
     "\n",
     "Have a look into the [llama-index documentation](https://docs.llamaindex.ai/en/stable/examples/) for more advanced RAG workflows."