|
15 | 15 | <div class="left box-gradient"> |
16 | 16 | <b>msg Local AI: Chat</b> is an application to locally run |
17 | 17 | Large Language Model (LLM) based generative Artificial |
18 | | - Intelligence (AI) characters (aka "chat-bots"). |
| 18 | + Intelligence (AI) characters (aka "chatbots"). |
19 | 19 | <p/> |
20 | 20 | It is based on the freely available <a href="https://faraday.dev">Faraday</a> LLM |
21 | 21 | host application, four pre-installed Open Source |
22 | 22 | <a href="https://mistral.ai/news/announcing-mistral-7b/">Mistral 7B</a> LLMs, and 24 |
23 | 23 | pre-configured Faraday AI characters. |
24 | 24 | <p/> |
25 | | - Each Faraday AI character is a |
26 | | - combination of a pre-selected LLM (fine-tuned for instruct or chat use), |
| 25 | + Each Faraday AI character combines |
| 26 | + a pre-selected LLM (fine-tuned for instruct or chat use), |
27 | 27 | pre-configured LLM parameters, and a corresponding pre-selected LLM. |
28 | 28 | </div> |
29 | 29 | <div class="right box-light centered"> |
|
46 | 46 | </a> |
47 | 47 |
|
48 | 48 | <p class="left-aligned"/> |
49 | | - The <b>msg Local AI: Chat</b> setup program does not require any |
| 49 | + The <b>msg Local AI: Chat</b> setup program does not require |
50 | 50 | administrator privileges during installation. Everything is installed |
51 | 51 | into <tt>%localappdata%</tt> and <tt>%appdata%</tt> only. A simple |
52 | 52 | click on the desktop icon "Uninstall msg Local AI Chat" will |
|
55 | 55 | </div> |
56 | 56 | <div class="row"> |
57 | 57 | <div class="left box-gradient fn-small"> |
58 | | - <b>Notice on STATISTICS</b>: <b>msg Local AI: Chat</b> uses the <i>Faraday</i> LLM host |
| 58 | + <b>Notice on USAGE STATISTICS</b>: <b>msg Local AI: Chat</b> uses the <i>Faraday</i> LLM host |
59 | 59 | application. <i>Faraday</i> sends anonymous usage information to its vendor. |
60 | 60 | <i>msg Research</i> explicitly checked this communication in 2024/02 and |
61 | | - considers it harmless from a GDPR point of view, as no personal |
| 61 | + considered it harmless from a GDPR point of view, as no personal |
62 | 62 | information were contained in the sent data. |
63 | 63 | <p/> |
64 | | - <b>Notice on CLOUD</b>: <b>msg Local AI: Chat</b> uses the <i>Faraday</i> LLM host |
| 64 | + <b>Notice on CLOUD ACCOUNT</b>: <b>msg Local AI: Chat</b> uses the <i>Faraday</i> LLM host |
65 | 65 | application. <i>Faraday</i> provides the possibility to create an account in |
66 | 66 | the Cloud on <tt>faraday.dev</tt>, login to it from within the application, |
67 | 67 | and use this account to upload and share chats and AI characters via |
68 | | - the Cloud. For the intended use of <b>msg Local AI: Chat</b> none of this is |
69 | | - functionality required or even recommended. The user is in charge when |
70 | | - using this additional functionality of <i>Faraday</i>. |
| 68 | + the Cloud. For the intended use of <b>msg Local AI: Chat</b>, none of this |
| 69 | + functionality is required or even recommended by msg Research. <b>msg Local AI: Chat</b> |
| 70 | + is intended to be used with the pre-defined AI characters only. |
| 71 | + The user is in charge when using the additional functionality of <i>Faraday</i>. |
71 | 72 | <p/> |
72 | 73 | <b>Notice on CONTENT</b>: <b>msg Local AI: Chat</b> uses <i>uncensored</i> LLMs which were |
73 | 74 | trained by <i>Mistral AI</i> with huge amounts of unknown, publicly available |
74 | | - data from the Internet. Hence the AI characters responses contain all |
75 | | - type of public information and in an entirely <i>unfiltered</i> format. The |
76 | | - user is responsible to treat the responses with care and especially |
77 | | - always consider and evaluate the responses in the context of the current |
78 | | - law and current social morality. |
| 75 | + data from the Internet. Hence, the AI character's responses contain all |
| 76 | + types of public information and in an entirely <i>unfiltered</i> format. The |
| 77 | + user is responsible for treating the responses with great care, validating them, |
| 78 | + and especially always considering and evaluating the responses in the context of the current |
| 79 | + law and social morality. |
79 | 80 | <p/> |
80 | | - <b>Notice on QUALITY</b>: <b>msg Local AI: Chat</b> is based on the decent but rather |
81 | | - small Open-Source LLM <i>Mistral 7B</i> in order to run even on computers |
| 81 | + <b>Notice on INTENDED USE-CASES</b>: <b>msg Local AI: Chat</b> provides 22 AI characters |
| 82 | + for the primary non-dialog-based use-cases. In addition, 2 general-purpose AI |
| 83 | + characters are provided (explicitly tagged as "GP") for the secondary dialog-based use-cases. |
| 84 | + The latter AI characters are provided with the intention of learning Prompt Engineering only. |
| 85 | + Please do not consider them as true alternatives to powerful general-purpose chatbots like OpenAI ChatGPT. |
| 86 | + <p/> |
| 87 | + <b>Notice on RESPONSE QUALITY</b>: <b>msg Local AI: Chat</b> is based on the decent but rather |
| 88 | + small Open-Source LLM <i>Mistral 7B</i> to run even on computers |
82 | 89 | with very constrained RAM and CPU resources. As a consequence, the |
83 | 90 | overall response quality obviously is noticeably reduced in contrast |
84 | | - to state-of-the-art chat-bots, like the commercial <i>GPT-3.5/GPT-4</i> based |
| 91 | + to state-of-the-art chatbots, like the commercial <i>GPT-3.5/GPT-4</i>-based |
85 | 92 | <i>OpenAI ChatGPT</i>. |
86 | 93 | </div> |
87 | 94 | <div class="right box-light"> |
|
99 | 106 | leveraging from generic text-based generative AI use-cases under full |
100 | 107 | offline-first and GDPR-compliance aspects. |
101 | 108 | <p/> |
102 | | - During installation the installer will download between 4GB and 16GB |
| 109 | + During installation, the installer will download between 4GB and 16GB |
103 | 110 | of LLM data from the <i>HuggingFace</i> portal, depending on the particular |
104 | 111 | features you selected during installation. |
105 | 112 | <p/> |
106 | | - <b>NOTICE</b>: Please be patient, the download will require many minutes, depending |
| 113 | + <b>NOTICE</b>: Please be patient; the download will require many minutes, depending |
107 | 114 | on how Internet connectivity. In case it breaks because of timeouts, just |
108 | 115 | repeat the step. |
109 | 116 | </div> |
110 | 117 | </div> |
111 | 118 | <div class="row"> |
112 | 119 | <div class="left box-gradient"> |
113 | | - After installation, you find an icon named "msg Local AI Chat" on your |
| 120 | + After installation, you will find an icon named "msg Local AI Chat" on your |
114 | 121 | desktop to run the <i>Faraday</i> LLM host application. After a short opt-in |
115 | 122 | dialog of the <i>Faraday</i> vendor, you can select the particular |
116 | 123 | AI character. |
117 | 124 | <p/> |
118 | 125 | You have 24 pre-defined AI characters available. |
119 | | - Each AI character has a name based on its use-case and the language |
120 | | - is speaks below its name. Click on the AI character to open a dialog |
| 126 | + Each AI character has a name based on its use-case, and the language |
| 127 | + it speaks below its name. Click on the AI character to open a dialog |
121 | 128 | with it. |
122 | 129 | </div> |
123 | 130 | <div class="right box-dark"> |
|
143 | 150 | Only the two "GP" (general purpose) AI characters named |
144 | 151 | "Assistant" are chat-based and hence accept full question/answer |
145 | 152 | dialog sessions. But beware of the above notice: Mistral 7B is |
146 | | - an excellent small LLM, but cannot fully compete with ChatCPT, of course! |
| 153 | + an excellent small LLM but cannot fully compete with ChatCPT, of course! |
147 | 154 | <p/> |
148 | 155 | <b>NOTICE</b>: just be patient on your initial request after |
149 | 156 | entering an AI character, as the underlying |
|
168 | 175 | <p class="left-aligned"/> |
169 | 176 | Many thanks to <i>Ahoy Labs, Inc.</i> for their awesome <i>Faraday</i> LLM host application |
170 | 177 | and to <i>Mistral AI</i> for their excellent <i>Mistral 7B</i> LLM. Your work |
171 | | - allows others to leverage from useful AI aspects without large constraints. |
| 178 | + allows others to leverage useful AI aspects without large constraints. |
172 | 179 | </div> |
173 | 180 | </div> |
174 | 181 | <div class="row"> |
|
0 commit comments