A different method to create AI
Year: 2026
Type: article
Abstract: Hi,I would like to share a project that I have dedicated a significant amount of time to.I have developed an approach that could potentially revolutionize the field of machine learning. As I understand it, the fundamental concept of machine learning is teaching a machine that, given a specific input (or type of input), it should produce a corresponding output. I believe there is an alternative method—one that is far more cost-effective than current approaches. I have spent a great deal of time exploring this method, which could also contribute to protecting our planet from environmental degradation. Here is a simplified breakdown of the approach:# Training Phase:Step 1:Tokenize all text using BPE (Byte Pair Encoding).Step 2:Assign a vector to each token, ensuring that words with similar spellings or meanings are positioned close to one another in the vector space.Step 3:Take your query or input, tokenize it, and convert each token into a fixed-length vector. Then, sum the columns of all these vectors to create a single, consolidated vector.Step 4:Use an unsupervised machine learning algorithm to categorize this final vector. Ensure that the distance between different categories is sufficiently large to avoid ambiguity. Step 5:Assign a category to each token. When an input falls into a specific category, the system identifies the corresponding token for that category as the "next token." This mapping is established during the training phase, allowing the system to generate outputs when similar inputs are provided later.Note: This is a simplified overview; you may encounter some challenges if you attempt to implement this directly. I likely have potential solutions for those issues, so please feel free to reach out if anything is unclear or if you have any questions.
Author: Sayed Anower
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236565110
