Merge pull request #86 from john0isaac/main

fix: typos and grammar 00-06

01-introduction-to-genai/README.md

@@ -4,7 +4,7 @@
 
 *(Click the image above to view video of this lesson)*
 
-Generative AI is artificial intelligence capable of generating text, images and other type of content. What makes it a fantastic technology is that it democratizes AI, anyone can use it with as little as a text prompt, a sentence written in a natural language. There's no need for you to learn a language like Java or SQL to accomplish something worthwhile, all you need is to use your language, state what you want and out comes a suggestion from an AI model. The applications and impact for this is huge, you write or understand reports, write applications and much more, all in seconds. 
+Generative AI is artificial intelligence capable of generating text, images and other types of content. What makes it a fantastic technology is that it democratizes AI, anyone can use it with as little as a text prompt, a sentence written in a natural language. There's no need for you to learn a language like Java or SQL to accomplish something worthwhile, all you need is to use your language, state what you want and out comes a suggestion from an AI model. The applications and impact for this is huge, you write or understand reports, write applications and much more, all in seconds. 
 
 In this curriculum, we’ll explore how our startup leverages generative AI to unlock new scenarios in the education world and how we address the inevitable challenges associated with the social implications of its application and the technology limitations.
 
@@ -51,15 +51,15 @@ A turning point arrived during the 90s, with the application of a statistical ap
 
 ### Neural networks and modern virtual assistants
 
-In more recent times, the technological evolution of the hardware, capable to handle larger amount of data and more complex computations, encouraged research in the AI fields, leading to the development of advanced machine learning algorithms – called neural networks or deep learning algorithms.
+In more recent times, the technological evolution of the hardware, capable of handling larger amounts of data and more complex computations, encouraged research in the AI fields, leading to the development of advanced machine learning algorithms – called neural networks or deep learning algorithms.
 
 Neural networks (and in particular Recurrent Neural Networks – RNNs) significantly enhanced natural language processing, enabling the representation of the meaning of text in a more meaningful way, valuing the context of a word in a sentence.
 
-This is the technology which powered the virtual assistants born in the first decade of the new century, very proficient in interpreting the human language, identifying a need, and performing an action to satisfy it – like answering with a pre-defined script or consuming a 3rd party service.
+This is the technology that powered the virtual assistants born in the first decade of the new century, very proficient in interpreting the human language, identifying a need, and performing an action to satisfy it – like answering with a pre-defined script or consuming a 3rd party service.
 
 ### Present day, Generative AI
 
-So that’s how we came to Generative AI today, that can be seen as a subset of deep learning.
+So that’s how we came to Generative AI today, which can be seen as a subset of deep learning.
 
 ![AI, ML, DL and Generative AI](./images/AI-diagram.png)
 
@@ -71,7 +71,7 @@ Most of the recent generative AI models – also known as Large Language Models
 
 In the next chapter we are going to explore different types of Generative AI models, but for now let’s have a look at how large language models work, with a focus on OpenAI GPT (Generative Pre-trained Transformer) models.
 
-* **Tokenizer, text to numbers**: Large Language Models receive a text as input and generate a text as output. However, being statistical models, they work much better with numbers than text sequences. That’s why every input to the model is processed by a tokenizer, before being used by the core model. A token is a chunk of text – consisting of a variable number of characters, so the tokenizer main task is splitting the input into an array of tokens. Then, each token is mapped with a token index, which is the integer encoding of the original text chunk.
+* **Tokenizer, text to numbers**: Large Language Models receive a text as input and generate a text as output. However, being statistical models, they work much better with numbers than text sequences. That’s why every input to the model is processed by a tokenizer, before being used by the core model. A token is a chunk of text – consisting of a variable number of characters, so the tokenizer's main task is splitting the input into an array of tokens. Then, each token is mapped with a token index, which is the integer encoding of the original text chunk.
 
 ![Example of tokenization](./images/tokenizer-example.png)
 
@@ -112,21 +112,21 @@ Also, the output of a generative AI model is not perfect and sometimes the creat
 
 Your assignment is to read up more on [generative AI](https://en.wikipedia.org/wiki/Generative_artificial_intelligence) and try to identify an area where you would add generative AI today that doesn't have it. How would the impact be different from doing it the "old way", can you do something you couldn't before, or are you faster? Write a 300 word summary on what your dream AI startup would look like and include headers like "Problem", "How I would use AI", "Impact" and optionally a business plan. 
 
-If you done this task, you might even be ready to apply to Microsoft's incubator, [Microsoft for Startups Founders Hub](https://www.microsoft.com/startups?WT.mc_id=academic-105485-koreyst) we offer credits for both Azure, OpenAI, mentoring and much more, check it out!
+If you did this task, you might even be ready to apply to Microsoft's incubator, [Microsoft for Startups Founders Hub](https://www.microsoft.com/startups?WT.mc_id=academic-105485-koreyst) we offer credits for both Azure, OpenAI, mentoring and much more, check it out!
 
 ## Knowledge check
 
 What's true about large language models?
 
 1. You get the exact same response every time.
 1. It does things perfectly, great at adding numbers, produce working code etc.
-1. The response may vary despite using the same prompt. It's also great at give you a first draft of something, be it text or code. But you need to improve on the results.
+1. The response may vary despite using the same prompt. It's also great at giving you a first draft of something, be it text or code. But you need to improve on the results.
 
 A: 3, an LLM is non-deterministic, the response vary, however, you can control its variance via a temperature setting. You also shouldn't expect it to do things perfectly, it's here to do the heavy-lifting for you which often means you get a good first attempt at something that you need to gradually improve.
 
 ## Great Work! Continue the Journey 
 
 Want to learn more about different Generative AI concepts? Go to the [continued learning page](../13-continued-learning/README.md?WT.mc_id=academic-105485-koreyst) to find other great resources on this topic. 
 
-Head over to the Lesson 2 where we will look at how to [explore and compare different LLM types](/02-exploring-and-comparing-different-llms/README.md?WT.mc_id=academic-105485-koreyst)! 
+Head over to Lesson 2 where we will look at how to [explore and compare different LLM types](/02-exploring-and-comparing-different-llms/README.md?WT.mc_id=academic-105485-koreyst)! 
 

02-exploring-and-comparing-different-llms/README.md

@@ -53,7 +53,7 @@ The term Foundation Model was [coined by Stanford researchers](https://arxiv.org
 Image source: [Essential Guide to Foundation Models and Large Language Models | by Babar M Bhatti | Medium
 ](https://thebabar.medium.com/essential-guide-to-foundation-models-and-large-language-models-27dab58f7404)
 
-To further clarify this distinction, let’s take ChatGPT as an example. To build the first version of ChatGPT, a model called GPT-3.5 served as the foundation model. This means that OpenAI used some chat-specific data to create a tuned version of GPT-3.5 that was specialized in performing well in conversational scenarios, such as chat bots.
+To further clarify this distinction, let’s take ChatGPT as an example. To build the first version of ChatGPT, a model called GPT-3.5 served as the foundation model. This means that OpenAI used some chat-specific data to create a tuned version of GPT-3.5 that was specialized in performing well in conversational scenarios, such as chatbots.
 
 ![Foundation Model](./images/Multimodal.png)
 
@@ -65,7 +65,7 @@ Another way to categorize LLMs is whether they are open source or proprietary.
 
 Open-source models are models that are made available to the public and can be used by anyone. They are often made available by the company that created them, or by the research community. These models are allowed to be inspected, modified, and customized for the various use cases in LLMs. However, they are not always optimized for production use, and may not be as performant as proprietary models. Plus, funding for open-source models can be limited, and they may not be maintained long term or may not be updated with the latest research. Examples of popular open source models include [Alpaca](https://crfm.stanford.edu/2023/03/13/alpaca.html), [Bloom](https://sapling.ai/llm/bloom) and [LLaMA](https://sapling.ai/llm/llama).
 
-Proprietary models are models that are owned by a company and are not made available to the public. These models are often optimized for production use. However, they are not allowed to be inspected, modified, or customized for different use cases. Plus, they are not always available for free, and may require a subscription or payment to use. Also, users do not have control over the data that is used to train the model, which means they should entrust the model owner with ensuring commitment about data privacy and responsible use of AI. Examples of popular proprietary models include [OpenAI models](https://platform.openai.com/docs/models/overview), [Google Bard](https://sapling.ai/llm/bard) or [Claude 2](https://www.anthropic.com/index/claude-2).
+Proprietary models are models that are owned by a company and are not made available to the public. These models are often optimized for production use. However, they are not allowed to be inspected, modified, or customized for different use cases. Plus, they are not always available for free, and may require a subscription or payment to use. Also, users do not have control over the data that is used to train the model, which means they should entrust the model owner with ensuring commitment to data privacy and responsible use of AI. Examples of popular proprietary models include [OpenAI models](https://platform.openai.com/docs/models/overview), [Google Bard](https://sapling.ai/llm/bard) or [Claude 2](https://www.anthropic.com/index/claude-2).
 
 ### Embedding versus Image generation versus Text and Code generation
 
@@ -99,7 +99,7 @@ Imagine that we can have someone as well who could create and review the quiz, t
 
 Now, let's talk about the difference between a service and a model. A service is a product that is offered by a Cloud Service Provider, and is often a combination of models, data, and other components. A model is the core component of a service, and is often a foundation model, such as an LLM.
 
-Services are often optimized for production use and are often easier to use than models, via a graphical user interface. However, services are not always available for free, and may require a subscription or payment to use, in exchange to leverage service owner’s equipment and resources, optimizing expenses and scaling easily. An example of service is [Azure OpenAI service](https://learn.microsoft.com/azure/ai-services/openai/overview?WT.mc_id=academic-105485-koreyst), which offers a pay-as-you-go rate plan,  meaning users are charged proportionally to how much they use the service Also, Azure OpenAI service offers enterprise-grade security and responsible AI framework on top of the models' capabilities.
+Services are often optimized for production use and are often easier to use than models, via a graphical user interface. However, services are not always available for free, and may require a subscription or payment to use, in exchange for leveraging the service owner’s equipment and resources, optimizing expenses and scaling easily. An example of service is [Azure OpenAI service](https://learn.microsoft.com/azure/ai-services/openai/overview?WT.mc_id=academic-105485-koreyst), which offers a pay-as-you-go rate plan,  meaning users are charged proportionally to how much they use the service Also, Azure OpenAI service offers enterprise-grade security and responsible AI framework on top of the models' capabilities.
 
 Models are just the Neural Network, with the parameters, weights, and others. Allowing companies to run locally, however, would need to buy equipment, build structure to scale and buy a license or use an open-source model. A model like LLaMA is available to be used, requiring computational power to run the model.
 
@@ -135,11 +135,11 @@ But when shall they consider fine-tuning a model rather than using a pre-trained
 
 There are several approaches a business can use to get the results they need from an LLM, you can select different types of models with different degrees of training
 
-deploy an LLM in production, with different levels of complexity, cost, and quality. Here's some different approaches:
+deploy an LLM in production, with different levels of complexity, cost, and quality. Here are some different approaches:
 
 - **Prompt engineering with context**. The idea is to provide enough context when you prompt to ensure you get the responses you need.
 
-- **Retrieval Augmented Generation, RAG**. Your data might exist in a database or web endpoint for example, to ensure this data, or a subset of it, is included at the time of prompting, you can fetch the relevant data and make that part of the users prompt.
+- **Retrieval Augmented Generation, RAG**. Your data might exist in a database or web endpoint for example, to ensure this data, or a subset of it, is included at the time of prompting, you can fetch the relevant data and make that part of the user's prompt.
 
 - **Fine-tuned model**. Here, you trained the model further on your own data which leads to the model being more exact and responsive to your needs but might be costly.
 
@@ -151,7 +151,7 @@ Img source: [Four Ways that Enterprises Deploy LLMs | Fiddler AI Blog](https://w
 
 Pre-trained LLMs work very well on generalized natural language tasks, even by calling them with a short prompt, like a sentence to complete or a question – the so-called “zero-shot” learning.
 
-However, the more the user can frame their query, with a detailed request and examples – the Context – the most accurate and closest to user’s expectations the answer will be. In this case, we talk about “one-shot” learning if the prompt includes only one example and “few shot learning” if it includes multiple examples.
+However, the more the user can frame their query, with a detailed request and examples – the Context – the more accurate and closest to user’s expectations the answer will be. In this case, we talk about “one-shot” learning if the prompt includes only one example and “few shot learning” if it includes multiple examples.
 Prompt engineering with context is the most cost-effective approach to kick-off with.
 
 ### Retrieval Augmented Generation (RAG)
@@ -168,7 +168,7 @@ This would be the preferred approach if:
 
 - **Using fine-tuned models**. A business would like to use fine-tuned less capable models (like embedding models) rather than high performance models, resulting in a more cost effective and fast solution.
 
-- **Considering latency**. Latency is important for a specific use-case, so it’s not possible to use very long prompts or the number of examples that should be learnt from the model doesn’t fit with the prompt length limit.
+- **Considering latency**. Latency is important for a specific use-case, so it’s not possible to use very long prompts or the number of examples that should be learned from the model doesn’t fit with the prompt length limit.
 
 - **Staying up to date**. A business has a lot of high-quality data and ground truth labels and the resources required to maintain this data up to date over time.
 
@@ -194,4 +194,4 @@ Read up more on how you can [use RAG](https://learn.microsoft.com/azure/search/r
 
 Want to learn more about different Generative AI concepts? Go to the [continued learning page](../13-continued-learning/README.md?WT.mc_id=academic-105485-koreyst) to find other great resources on this topic.
 
-Head over to the Lesson 3 where we will look at how to [build with Generative AI Responsibly](/03-using-generative-ai-responsibly/README.MD?WT.mc_id=academic-105485-koreyst)!
+Head over to Lesson 3 where we will look at how to [build with Generative AI Responsibly](/03-using-generative-ai-responsibly/README.MD?WT.mc_id=academic-105485-koreyst)!

03-using-generative-ai-responsibly/README.MD

@@ -48,7 +48,7 @@ The model produces a response like the one below:
 
 > *(Source: [Flying bisons](https://flyingbisons.com))*
 
-This is a very confident and thorough answer. Unfortunately, it is incorrect. Even with a minimal amount of research, one would discover there was more than one survivor of the Titanic survivor. For a student who is just starting to research this topic, this answer can be persuasive enough to not be questioned and treated as fact. The consequences of this can lead to AI system to be unreliably and negatively impact the reputation of our startup.
+This is a very confident and thorough answer. Unfortunately, it is incorrect. Even with a minimal amount of research, one would discover there was more than one survivor of the Titanic survivor. For a student who is just starting to research this topic, this answer can be persuasive enough to not be questioned and treated as fact. The consequences of this can lead to the AI system being unreliable and negatively impact the reputation of our startup.
 
 With each iteration of any given LLM, we have seen performance improvements around minimizing hallucinations. Even with this improvement,  we as application builders and users still need to remain aware of these limitations.
 
@@ -94,7 +94,7 @@ It is now time to find ways where we can prevent or limit the potential harm cau
 
 - **Safety System**. A safety system is a set of tools and configurations on the platform serving the model that help mitigate harm. An example of this is the content filtering system on the Azure OpenAI service. Systems should also detect jailbreak attacks and unwanted activity like requests from bots.
 
-- **Metaprompt**. Metaprompts and grounding are ways we can direct or limit the model based on certain behaviors and information. This could be using system inputs to define certain limits of the model. In addition, providing outputs that are more relevant to scope or domain of system.
+- **Metaprompt**. Metaprompts and grounding are ways we can direct or limit the model based on certain behaviors and information. This could be using system inputs to define certain limits of the model. In addition, providing outputs that are more relevant to the scope or domain of the system.
 
  It can also be using techniques like Retrieval Augmented Generation (RAG) to have the model only pull information from a selection of trusted sources. There is a lesson later in this course for [building search applications](../08-building-search-applications/README.md?WT.mc_id=academic-105485-koreyst)
 
@@ -120,7 +120,7 @@ What are some things you need to care about to ensure responsible AI usage?
 1. Harmful usage, that AI isn't used for criminal purposes.
 1. Ensuring the AI is free from bias and discrimination.
 
-A: 2 and 3 is correct. Responsible AI helps you consider how to mitigate harmful effects and biases and more.
+A: 2 and 3 are correct. Responsible AI helps you consider how to mitigate harmful effects and biases and more.
 
 ## 🚀 Challenge