From 444fc50acefb510d976041104016484a5bf80728 Mon Sep 17 00:00:00 2001
From: Andri Joos <andri@joos.io>
Date: Thu, 13 Jun 2024 21:32:48 +0200
Subject: [PATCH] fix assistant comparison
---
src/04_Appendix/assistant-comparison.tex | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/src/04_Appendix/assistant-comparison.tex b/src/04_Appendix/assistant-comparison.tex
index b428cb1..d15f3e5 100644
--- a/src/04_Appendix/assistant-comparison.tex
+++ b/src/04_Appendix/assistant-comparison.tex
@@ -8,7 +8,7 @@
The \nameref{sec:appendix/testing-protocols/llm-behavior-tests} clearly show, that the Mixtral assistant currently has some significant deficites.
In particular, the hallucination of ingredient quantities and the inability to provide step-by-step guidance make the assistant very unreliable.
- The OpenAI assistant however provides assistance in the most common use cases.
+ The OpenAI assistant on the other hand provides assistance in all common use cases.
\section{Processing Speed}\label{sec:appendix/assistant-comparison/processing-speed}
The processing speed is tested for both text input and audio input.
@@ -42,13 +42,14 @@
\end{figure}
\subsection{Interim Conclusion}
- The OpenAI assistant is faster than the Mixtral assistant in all aspects.
+ Although these times are not entirely representative, they clearly show that the OpenAI assistant is much faster than the Mixtral assistant in all aspects.
\section{Privacy}\label{sec:appendix/assistant-comparison/privacy}
- The Mixtral assistant offers a high level of privacy because the Mixtral LLM used is hosted at OST and therefore the user's input is not sent to an external service.
+ The Mixtral assistant offers a high level of privacy because the used Mixtral LLM is hosted at OST and therefore the user's input is not sent to an external service.
The conversation is also only stored for one day, after which it is deleted and cannot be recovered.
The OpenAI assistant on the other hand uses the OpenAI API.
+ OpenAI stores the conversation history forever.
However, OpenAI claims, that they \href{https://platform.openai.com/docs/introduction}{do not use API data for training}.
\subsection{Interim Conclusion}
@@ -56,4 +57,4 @@
\section{Conclusion}
The previous comparisons clearly show, that the OpenAI assistant performs better than the Mixtral assistant in both \hyperref[sec:appendix/assistant-comparison/behavior-tests]{response quality and correctness} and \hyperref[sec:appendix/assistant-comparison/processing-speed]{speed}.
- What is better about the Mixtral assistant is its privacy.
+ What is better about the Mixtral assistant is its \hyperref[sec:appendix/assistant-comparison/privacy]{privacy}.
--
GitLab