AI search is a joke – Exhibit 3

When you are looking for a precise sequence words, AI is not the appropriate tool. A precise search works very nicely. AI doesn't know texts - it knows networks. Wikipedia unfortunately has a technical article not real world - is https://vitalflux.com/deep-neural-network-examples-from-real-life/ understandable to you or is there too much mathematical jargon?

or does this answer from perplexity add clarity?

The difference between searching texts and searching a neural network built from texts lies in the fundamental approach to information retrieval and processing. These two methods represent distinct paradigms in natural language processing and information retrieval.

Text Search

Text search, also known as full-text search, involves directly querying a collection of documents or texts to find specific words, phrases, or patterns. This method typically uses techniques such as:

Inverted indexing: Creating an index that maps words to their locations in the documents

Boolean operators: Using AND, OR, NOT to combine search terms

Relevance ranking: Scoring documents based on term frequency and other factors

Text search is generally faster for exact matches and can be more precise when looking for specific information. However, it has limitations in understanding context, semantics, and relationships between concepts.

Neural Network Search

Searching a neural network built from texts, on the other hand, involves querying a model that has been trained on a large corpus of text data. This approach leverages the power of machine learning and natural language processing to understand and generate human-like responses. Key aspects include:

Semantic understanding: The ability to grasp the meaning and context of words and phrases

Latent space representation: Encoding text into high-dimensional vectors that capture semantic relationships

Transfer learning: Applying knowledge gained from training on a large corpus to specific tasks or domains

Neural network search can handle more complex queries, understand context, and even generate new text based on the input. It excels at tasks like question answering, summarization, and inferring relationships between concepts.

Key Differences

Representation: Text search works with raw text, while neural networks transform text into complex numerical representations.

Flexibility: Neural networks can handle more varied and nuanced queries, while text search is more rigid and literal.

Contextual understanding: Neural networks can grasp context and semantics, whereas text search is limited to pattern matching.

Computational requirements: Text search is generally faster and requires less computational power, while neural networks are more resource-intensive.

Accuracy vs. Generalization: Text search is more accurate for finding exact matches, while neural networks excel at generalizing and understanding implicit relationships.

Updating: Text search databases can be easily updated by adding or removing documents, while neural networks typically require retraining to incorporate new information.

In practice, many modern systems combine both approaches, leveraging the strengths of each method to provide more comprehensive and intelligent search capabilities.

I agree that misunderstandings as to what AI does and how to use it are very common among users - myself included. The results are that people are very disappointed with the results - not understanding that AI was not the correct tool for what they wanted. There are also people who are overly enthusiastic, because they discovered new, interesting information. Unfortunately, they don't realize that some of this new information is either false or unverified. Until most users understand that the AI does not deal with texts but with mathematical constructs based on many texts and the AI does not attempt exhaustive searches but only searches to the point of a reasonable answer, most users will abuse AI. My hope is that many of the kids currently in junior colleges, high school, grade schools will learn the appropriate use of AI … and when they become the majority of average users . . .

You need to reword questions in all the AI bot's I've tried. The pattern of requiring rewording implies that chat-bot Reinforcement Learning from Human Feedback (RLHF) still has trouble identifying the key words … it goes off on a tangent related to a secondary word. It is usually easy to identify the mistake and reword the question in a way that is clearer to the bot … but that is not a guarantee that further down in the conversation the bot will not revert to the erroneous understanding. In a conversation on Jewish divisions of Torah text in the 1 year and 3 year cycle, I had to give up. I could not find a bot that didn't slip into confusing the two. I also have trouble keeping any bot on track when I don't want it to limit itself to the Protestant canon.