docs: update some links and language; add LLM family tree because it'…

…s neat

.gitignore

@@ -2,4 +2,5 @@ __pycache__
 _build
 .ipynb_checkpoints
 *.sqlite
-.env
+.env
+.DS_Store

SentimentAnalysisWorkshop.ipynb

@@ -333,17 +333,41 @@
     "\n",
     "In this case we're taking advantage of an existing [language model](https://en.wikipedia.org/wiki/Language_model), VADER, that has been trained to analyze sentiment in text, but if we wanted to train our own model, it would be a much more involved process.\n",
     "\n",
-    "With the advent of [Large Language Models](https://en.wikipedia.org/wiki/Large_language_model) (LLMs), like the [Generative Pre-Trained Transformer](https://en.wikipedia.org/wiki/Generative_pre-trained_transformer) (GPT) models that power ChatGPT - and the various [other models that have exploded in popularity](https://informationisbeautiful.net/visualizations/the-rise-of-generative-ai-large-language-models-llms-like-chatgpt/) since - we can leverage the powerful inference and predictive capabilities of these models to perform tasks like sentiment analysis with greater accuracy without having to train our own models.\n",
+    "With the advent of [Large Language Models](https://en.wikipedia.org/wiki/Large_language_model) (LLMs), like the [Generative Pre-Trained Transformer](https://en.wikipedia.org/wiki/Generative_pre-trained_transformer) (GPT) models that power ChatGPT [large language models have exploded in popularity](https://informationisbeautiful.net/visualizations/the-rise-of-generative-ai-large-language-models-llms-like-chatgpt/).\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "da92b914",
+   "metadata": {},
+   "source": [
+    "### LLM family tree\n",
     "\n",
-    "We can even leverage some prompting techniques - which we'll explore in later cells - to quickly teach the model how to perform more unique analyses and refine our results.\n"
+    "<div style=\"display: flex; alight-items: center; justify-content: center;\"><a href=\"https://github.com/Mooler0410/LLMsPracticalGuide\" target=\"_blank\"><img alt=\"LLM Evolutionary Tree\" src=\"./assets/llm-family-tree.gif\" /><a/></div>\n",
+    "\n",
+    "This visualiztion from [Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond](https://arxiv.org/abs/2304.13712) provides a great overview of how language models have evolved over time and gives you a sense of just how much things have been developing in the last 12 months.\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "fabe668a",
+   "metadata": {},
+   "source": [
+    "### The power of LLMs\n",
+    "\n",
+    "We can leverage the inference and predictive capabilities of these models to perform tasks like sentiment analysis with greater accuracy without having to train our own models.\n",
+    "\n",
+    "We can even leverage some prompting techniques - which we'll explore in later cells - to quickly teach the model how to perform more unique analyses and refine our results.\n",
+    "\n",
+    "In the past, these would have been a significant undertaking, but now we can acheive similar results with some simple prompting.\n"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "7c607a30",
    "metadata": {},
    "source": [
-    "## Real world example\n",
+    "## Real world data\n",
     "\n",
     "Let's take a look at how this works with text generated by other humans (_probably_) without expecting someone would be trying to analyze the sentiment of their text.\n",
     "\n",
@@ -474,7 +498,15 @@
     "\n",
     "In responding to our prompts, ChatGPT follows a similar process to the NLP workflow described above.\n",
     "\n",
-    "It breaks our prompts into tokens, predicts which tokens should logically follow the ones that we've provided, and returns that text.\n",
+    "It breaks our prompts into [tokens](https://learn.microsoft.com/en-us/semantic-kernel/prompt-engineering/tokens), predicts which tokens should logically follow the ones that we've provided, and returns that text.\n",
+    "\n",
+    "ChatGPT's tuning based on Reinforcement Learning from Human Feedback ([RLHF](https://www.assemblyai.com/blog/how-rlhf-preference-model-tuning-works-and-how-things-may-go-wrong/)) is what lead it to be so popular, and is also part of what makes it so powerful.\n",
+    "\n",
+    "#### Learn more\n",
+    "\n",
+    "- [How ChatGPT Actually Works](https://www.assemblyai.com/blog/how-chatgpt-actually-works/)\n",
+    "- [How ChatGPT Works: The Models Behind The Bot](https://towardsdatascience.com/how-chatgpt-works-the-models-behind-the-bot-1ce5fca96286)\n",
+    "- [The inside story of how ChatGPT was built from the people who made it](https://www.technologyreview.com/2023/03/03/1069311/inside-story-oral-history-how-chatgpt-built-openai/)\n",
     "\n",
     "### Tokens\n",
     "\n",
@@ -571,9 +603,13 @@
     "- **User**: user messages are the individual prompts that the user sends to the model\n",
     "- **Assistant**: assistant messages are the responses the model generates to the user's prompts\n",
     "\n",
+    "If you're just chatting with ChatGPT via it's web-based User Interface (UI), you're probably familiar with **User** and **Assistant** messages, but you may not know that there's a **System** message behind the scenes that helps guide how the model responds to your messages.\n",
+    "\n",
+    "[Custom Instructions](https://openai.com/blog/custom-instructions-for-chatgpt) are sort of like [system prompts](https://github.com/jujumilk3/leaked-system-prompts), but don't give us quite as much control as we can exercise via the [Chat API](https://platform.openai.com/docs/api-reference/chat) - or as much control as we can get with some of the recent open source models.\n",
+    "\n",
     "### Example conversation document\n",
     "\n",
-    "The whole thing looks a bit like this:\n",
+    "When you put it all together, the whole thing looks a bit like this:\n",
     "\n",
     "```\n",
     "[System]\n",