AI Explains AI (part four)

Bing Chat on Bing Chat

This series of articles are some introductory questions and high-level answers posed to various AI technologies to uncover how they work, ‘in their own words’. Some questions were quite naive at first, in part due to my initial lack of detailed technical understanding of how AI tools work under the covers (haven’t seen the source code or talked to the engineers), and being perfectly willing to let the AI to correct the premise, which is partially enlightening in itself, and reveal technical facts of how they actually work.

This particular one was problematic — note that the first three answers Bing gave were identical despite changing the phrasing of the question, and further answers often appeared to be a summary of the results of a typical web search, rather than composing a response based on the prompt. Twice Bing cut me off saying I’d reached the limit, once I had to sign in, then it cut me off again, with no recourse to continue (see the last Bing response at the end). Despite all that, there is valuable information in the responses.

This conversation was conducted with Bing Chat stand-alone version (not the Edge browser version) as of October, 2023, using Precise mode.

Q. When generating step by step instructions for cooking filet mignon, how do you determine which words or sequences of words are appropriate for the response?

Bing Chat:

When generating step-by-step instructions for cooking filet mignon, it is important to use clear and concise language that is easy to understand. Use action verbs to describe each step, and make sure the sequence of steps is logical and easy to follow. It’s also helpful to use visual aids such as images or videos to supplement the text instructions.

When deciding which words or sequences of words are appropriate for the response, consider the following:

• Audience: Who is the target audience? Are they experienced cooks or beginners?
• Purpose: What is the purpose of the instructions? Are they meant to be a quick reference guide or a comprehensive tutorial?
• Context: What is the context of the instructions? Are there any specific tools, ingredients, or techniques that need to be explained?

To get started with generating step-by-step instructions for cooking filet mignon, you can search for existing recipes online and analyze how they are written. For example, you can check out this video by Max the Meat Guy on YouTube or this recipe by Delish. You can also use your own experience and knowledge to create your own set of instructions.

I hope this helps!

1. https://www.youtube.com/watch?v=2yUUmkYHOJE
2. https://www.youtube.com/watch?v=psp702ghkaI
3. https://www.youtube.com/watch?v=fDrkI9_EEe8
4. https://www.delish.com/cooking/recipe-ideas/a23515113/how-to-cook-filet-mignon/
5. https://thebigmansworld.com/filet-mignon-recipe/
6. https://nicksgrillasheville.com/how-to-cook-filet-mignon-on-blackstone-griddle/
7. https://tasty.co/recipe/filet-mignon
8. https://www.instagram.com/thatdude_cancook/
9. https://www.tiktok.com/@thatdudecancook?lang=en

Q. I’m not asking for general instructions. When generating step by step instructions for cooking filet mignon, how exactly does Bing Chat determine which words or sequences of words are appropriate for the response?

Bing Chat:

When generating step-by-step instructions for cooking filet mignon, it is important to use clear and concise language that is easy to understand. Use action verbs to describe each step, and make sure the sequence of steps is logical and easy to follow. It’s also helpful to use visual aids such as images or videos to supplement the text instructions.

When deciding which words or sequences of words are appropriate for the response, consider the following:

• Audience: Who is the target audience? Are they experienced cooks or beginners?
• Purpose: What is the purpose of the instructions? Are they meant to be a quick reference guide or a comprehensive tutorial?
• Context: What is the context of the instructions? Are there any specific tools, ingredients, or techniques that need to be explained?

To get started with generating step-by-step instructions for cooking filet mignon, you can search for existing recipes online and analyze how they are written. For example, you can check out this video by Max the Meat Guy on YouTube or this recipe by Delish. You can also use your own experience and knowledge to create your own set of instructions.

I hope this helps!

Bing Chat suggested follow-up query - “How does Bing Chat generate step-by-step instructions?”

Bing Chat:

When generating step-by-step instructions for cooking filet mignon, it is important to use clear and concise language that is easy to understand. Use action verbs to describe each step, and make sure the sequence of steps is logical and easy to follow. It’s also helpful to use visual aids such as images or videos to supplement the text instructions.

When deciding which words or sequences of words are appropriate for the response, consider the following:

• Audience: Who is the target audience? Are they experienced cooks or beginners?
• Purpose: What is the purpose of the instructions? Are they meant to be a quick reference guide or a comprehensive tutorial?
• Context: What is the context of the instructions? Are there any specific tools, ingredients, or techniques that need to be explained?

To generate step-by-step instructions for cooking filet mignon, you can search for existing recipes online and analyze how they are written. You can also use your own experience and knowledge to create your own set of instructions.

I hope this helps!

Q. How does Bing Chat choose appropriate words for responses?

Bing Chat:

When generating responses, Bing Chat uses a combination of artificial intelligence models and natural language processing techniques to understand the user’s message and generate an appropriate response. The specific algorithms used depend on the context of the conversation and the user’s language.

Bing Chat also takes into account the mode it is currently in. The available modes are Balanced, Creative, and Precise. In Balanced mode, Bing Chat aims to provide responses that are reasonable and coherent, balancing accuracy and creativity in conversation. In Creative mode, Bing Chat generates longer, more imaginative answers, while in Precise mode, it focuses on shorter, factual answers (1, 2, 3).

When deciding which words or sequences of words are appropriate for the response, Bing Chat considers various factors such as the audience, purpose, and context of the conversation. For example, if the user is asking for step-by-step instructions for cooking filet mignon, Bing Chat would use clear and concise language with action verbs to describe each step(4).

I hope this helps!

1. https://www.zdnet.com/article/how-to-use-the-new-bing-and-how-its-different-from-chatgpt/
2. https://bgr.com/tech/you-can-now-tell-bing-to-either-be-more-entertaining-or-precise-when-it-talks-to-you/
3. https://www.thurrott.com/cloud/280006/microsoft-bing-chatbot-conversation-style-selector
4. https://nerdschalk.com/what-are-bings-conversation-styles-and-how-do-they-work/
5. https://www.techrepublic.com/article/bing-chat-ai-tips-for-better-results/

Q. Please explain in exacting detail what artificial intelligence models and natural language processing techniques are and precisely how they work. Do not summarize or use generalizations.

Bing Chat:

Explore (https://en.wikipedia.org/wiki/Natural_language_processing)

Artificial Intelligence (AI) models are programs that have been trained on a set of data to recognize certain patterns or make certain decisions without further human intervention1. They apply different algorithms to relevant data inputs to achieve the tasks, or output, they’ve been programmed for1.

AI models can be categorized into two types: rule-based AI models and Machine Learning (ML) models1. Rule-based AI models, also known as rules engines, expert systems, knowledge graphs or symbolic AI, are a series of if-then-else statements, with rules programmed explicitly by a data scientist1. On the other hand, ML models use statistical AI rather than symbolic AI. These models are “trained” by applying their mathematical frameworks to a sample dataset whose data points serve as the basis for the model’s future real-world predictions1.

ML model techniques can generally be separated into three broad categories: supervised learning, unsupervised learning and reinforcement learning1. In supervised learning, a human expert labels training data. For example, a data scientist training an image recognition model to recognize dogs and cats must label sample images as “dog” or “cat”, as well as key features — like size, shape or fur — that inform those primary labels1.

Natural Language Processing (NLP) is a subfield of AI that gives computers the ability to understand and manipulate human language. NLP combines computational linguistics — rule-based modeling of human language — with statistical, machine learning, and deep learning models2.

There are several techniques used in NLP:

1. Tokenization: This is one of the most primary and simple NLP techniques when doing natural language processing. It involves breaking down text into words or other meaningful elements called tokens3.
2. Stemming and Lemmatization: These techniques are used to reduce inflectional forms and sometimes derivationally related forms of a word to a common base form3.
3. Stop Words Removal: Stop words are a set of commonly used words in any language. The intuition behind using stop words is that these words do not carry much meaning and are usually removed from texts3.
4. TF-IDF: Term Frequency-Inverse Document Frequency (TF-IDF) is a numerical statistic used to reflect how important a word is to a document in a collection or corpus3.
5. Keyword Extraction: This technique is used to extract important and relevant terms from the document3.
6. Word Embeddings: This is the collective name for a set of language modeling and feature learning techniques in NLP where words or phrases from the vocabulary are mapped to vectors of real numbers3.
7. Sentiment Analysis: This technique is used to identify and extract subjective information from source materials4.
8. Named Entity Recognition: This technique is used to locate and classify named entities in text into pre-defined categories such as person names, organizations, locations, medical codes, time expressions, quantities, monetary values, percentages, etc4.
9. Summarization: This technique involves reducing large texts into their most important parts4.
10. Topic Modeling: This technique is a type of statistical model for discovering the abstract “topics” that occur in a collection of documents4.

Each of these techniques plays a crucial role in understanding and interpreting human language in a way that machines can understand.

Sidebar links:

A. The first practical application of Natural Language Processing was the translation of the messages from Russian to English to understand what the commies were at. (sic)(https://theappsolutions.com/blog/development/natural-language-processing/#:~:text=The%20first%20practical%20application%20of%20Natural)

B. The first major leap forward for natural language processing algorithm came in 2013 with the introduction of Word2Vec — a neural network based model used exclusively for producing embeddings. (https://www.articlecity.com/blog/natural-language-processing-algorithms-nlp-ai/#:~:text=The%20first%20major%20leap%20forward%20for)

C. ABSTRACT Natural Language Processing is a technique where mechine (sic) can become more human and there by reducing the distance between human being and the mechine (sic) can bereduced (sic). (https://www.engpaper.com/natural-language-processing.htm#:~:text=ABSTRACT%20Natural%20Language%20Processing%20is%20a)

Footnotes

1. https://www.ibm.com/topics/ai-model
2. https://www.ibm.com/topics/natural-language-processing
3. https://www.projectpro.io/article/10-nlp-techniques-every-data-scientist-should-know/415
4. https://monkeylearn.com/blog/natural-language-processing-techniques/
5. https://en.wikipedia.org/wiki/Artificial_intelligence
6. https://bing.com/search?q=artificial+intelligence+models
7. https://viso.ai/deep-learning/ml-ai-models/
8. https://thedecisionlab.com/reference-guide/computer-science/artificial-intelligence-models
9. https://en.wikipedia.org/wiki/Natural_language_processing
10. https://builtin.com/data-science/introduction-nlp
11. https://www.gettyimages.com/detail/illustration/natural-language-processing-concept-business-royalty-free-illustration/1193264709

Q. Please explain in exacting detail what Bing Chat’s artificial intelligence models and Bing Chat’s natural language processing techniques are and precisely how Bing Chat’s models and techniques work in detail. Do not summarize or use generalizations.

Bing Chat:

Bing Chat uses a combination of AI models and NLP techniques to understand and respond to user queries. Here’s a detailed explanation:

1. Language Understanding: Bing Chat uses a transformer-based language model, which is a type of AI model that processes text data. It’s trained on a large corpus of text data and can generate human-like text based on the input it receives. The transformer model architecture allows the model to pay varying amounts of attention to different words in the input sequence when generating output, making it highly effective for many NLP tasks.
2. Text Generation: Bing Chat uses a sequence-to-sequence (Seq2Seq) model for generating responses. Seq2Seq is a method in machine learning where a model is trained to convert sequences from one domain (e.g., sentences in English) into sequences in another domain (e.g., the corresponding sentences in French). In the case of Bing Chat, the input sequence is the user’s query, and the output sequence is the assistant’s response.
3. Tokenization: Bing Chat uses tokenization to break down user queries into individual words or tokens. This is an essential step in NLP as it allows the model to understand and process each word individually.
4. Named Entity Recognition (NER): Bing Chat uses NER to identify important elements in the user’s query, such as names of people, places, organizations, dates, etc. This helps the model understand the context of the query better.
5. Sentiment Analysis: Bing Chat uses sentiment analysis to understand the emotional tone behind the user’s query. This helps in generating responses that are appropriate to the user’s emotional state.
6. Text Summarization: Bing Chat uses text summarization techniques to generate concise summaries of long pieces of text. This is particularly useful when providing brief summaries of search results or long articles.
7. Topic Modeling: Bing Chat uses topic modeling to identify the main topics present in a piece of text. This helps in understanding the overall theme or subject of the user’s query.
8. Web Search: Bing Chat has access to a web search tool that allows it to search the web for information relevant to the user’s query. The search results are then used to generate informative and accurate responses.
9. Image Generation: Bing Chat also has access to an image generation tool that allows it to create graphical artwork based on prompts provided by users.

These models and techniques work together seamlessly to provide users with accurate, informative, and engaging responses.

Q. Precisely how does Bing Chat determine the audience, purpose, and context of the conversation? Explain in exact detail, do not summarize or use generalizations.

Bing Chat:

Bing Chat uses a combination of techniques to determine the audience, purpose, and context of the conversation:

1. Audience: Bing Chat assumes that the audience is the user who is currently interacting with it. It doesn’t have access to personal data about users unless explicitly provided in the course of the conversation. It uses the content and language of the user’s messages to make inferences about the user’s knowledge level, interests, and intent.
2. Purpose: The purpose of each conversation with Bing Chat is determined by the user’s inputs. Users interact with Bing Chat for various reasons, such as asking questions, seeking information, requesting for tasks to be performed, or simply chatting. Bing Chat aims to understand and fulfill these user intents to the best of its ability.
3. Context: The context of the conversation includes both the immediate conversational context and any available metadata. The immediate conversational context consists of the current turn and past turns in the same conversation. Bing Chat uses this context to maintain coherence across turns and to refer back to previous user inputs where necessary. Metadata can include information like the time at which the conversation started and the user’s location, which can be used to provide more relevant responses.

In terms of choosing words and phrases for its responses, Bing Chat uses a language model trained on a diverse range of internet text. However, it doesn’t know specifics about which documents were in its training set or have access to any personal data about individuals unless explicitly provided in the course of the conversation.

The language model generates responses word by word, starting from an initial prompt that includes both the user’s message and some additional instructions or guidance. At each step, it uses statistical patterns learned during training to predict the next word that’s most likely to follow given the current context (the words generated so far). This process is repeated until a complete response is generated.

The exact phrasing of Bing Chat’s responses is influenced by a variety of factors, including:

• The specific words used in the user’s message.
• The conversational context (e.g., previous messages in the conversation).
• High-level guidance encoded in its initial prompt.
• Randomness inherent in its generation process (it’s probabilistic, not deterministic).
• Fine-tuning on specific dialogue datasets to learn conversational patterns.
• Regular updates and improvements based on user feedback and other data.

It’s important to note that while Bing Chat strives for accuracy and relevance in its responses, it doesn’t have beliefs or opinions. Any claims it makes are based on patterns in the data it was trained on.

Q. What are the specific differences between ChatGPT, Bing Chat, Bard and Claude? Please elaborate in detail, do not use generalizations or summarize.

Bing Chat:

Here are the specific differences between ChatGPT, Bing Chat, Bard, and Claude:

1. ChatGPT: Developed by OpenAI, ChatGPT is a large language model-based chatbot that stands out for its ability to steer and refine conversations according to user preferences (1). It allows users to dictate the length, format, style, level of detail, and language used in the conversation (1). This is achieved through a process known as prompt engineering, where successive prompts and replies are considered at each conversation stage as a context (1). The latest versions, ChatGPT-3.5 and GPT-4, are fine-tuned for conversational applications using a combination of supervised and reinforcement learning techniques (1).
2. Bing Chat: Bing Chat is an AI-powered version of Microsoft’s Bing search engine (2). It uses an upgraded version of ChatGPT’s technology (2). Once inside, queries are currently capped to 150 per day, with 15 back-and-forth exchanges each time (2).
3. Bard: Google Bard is an experimental AI chat service powered by the Pathways Language Model 2 (PaLM 2) (1). Although still under development, Bard has shown promise in generating text, translating languages, creating various types of creative content, and providing informative answers to user queries (1).
4. Claude: Claude 2.0 is a large language model from Anthropic (3). It excels at summarizing documents but tends to hallucinate answers, making it unsuitable for data analysis (1).

Each of these AI models brings unique features and capabilities to the table. The choice of which one to use depends on your specific needs and preferences (2, 1).

1. https://www.geeky-gadgets.com/chatgpt-vs-bing-vs-bard-vs-claude/
2. https://www.pcmag.com/news/chatgpt-vs-google-bard-vs-microsoft-bing-which-ai-chatbot-gives-best-answers
3. https://annmichaelsen.com/2023/08/28/bard-vs-chatgpt-vs-bing-vs-claude-the-right-llm-for-every-task/
4. https://sourceforge.net/software/compare/Bard-vs-Bing-Chat-vs-ChatGPT-vs-Claude/
5. https://www.techmehow.com/chatgpt-vs-bing-vs-bard-vs-claude-comparison/

Bing Chat:

Sorry, this conversation has reached its limit. Use the “broom” button to sweep this away and chat more.

(Author’s note: I don’t see a “broom” button anywhere on the interface. Do you?)

Previous in the series: Claude on Claude

Afterword

Some additional reading on Bing Chat and how it works.

Bing Chat — Microsoft Edge version: https://www.microsoft.com/en-us/edge/features/bing-chat?form=MT00D8

Microsoft forum entry about Bing Chat being changed to Bing Discover: https://answers.microsoft.com/en-us/microsoftedge/forum/all/microsoft-edge-suddenly-not-offering-bing-chat/8cb54cc5-1012-44c0-b6dc-92909a7974c2

Third party information about usage of Bing Chat, stand-alone version: https://www.zdnet.com/article/how-to-use-the-new-bing-and-how-its-different-from-chatgpt/