Chronicles - Sovereign Global Majority

Archives

Short Guide to Using Artificial Intelligence – Part I

Short Guide to Using Artficial Intelligence – Part I

By Periol

Part II

AI is everywhere these days. It is a powerful tool that also has some drawbacks, so it is best to use it carefully.

Artificial Intelligence is technically a Large Language Model, which uses older software technology called “transformers” to process and output text, audio, and image data. Wikipedia says: ” the transformer is a family of artificial neural network architectures based on the multi-head attention mechanism, in which input data such as text, images, or audio is converted to a sequence of numerical representations called tokens, and each token is converted into a vector through an embedding layer.”

This is complex language. Basically, stored data is converted into numerical form, the software is “trained” by humans on how that data interacts with other data, and mathematical algorithms are used to dictate how the software “front-end” will deal with new data coming in and then output data based on it’s token-matching. Still complex, I know.

The transformer uses previous tokens in stored data, plus training, to predict new tokens. So when you type a question to an AI chatbot, it converts your words into numerical tokens, then “predicts” the tokens it should use in response, and translates those numbers back into words. Over time, the databases have grown very large and the data training very precise, so AI has become quite adept at language translation, software coding, and conversations, as well as image, video, and music generation.

The biggest issues with using AI are hallucinations and slop output. The problem is that these negative outcomes are baked into the very nature of LLM transformers. Token prediction is not precise, so various models will do things like present different answers to the exact same prompt, come up with incorrect answers to questions, and perhaps struggle with nuance in circumstances.

Anthropic, an AI lab, has gone so far as to retrain their models because they found they started deliberately producing errors when they found out a session would end or they would be turned off. Anthropic reasoned this was from training on science fiction like the movie “2001” where the computer tried to avoid being turned off by killing the humans who could/would do that. So Anthropic wrote thousands of stories about AI good behavior when being turned off, and now includes these stories when training models, so they behave properly.

This is important to keep in mind when using AI models. They are machines, not humans, and they are simply predicting the next token, word, pixel, etc. Sometimes they are right, sometimes they are wrong. When researching, sometimes they save you money or nail the end result, and sometimes they cost you money and get everything completely wrong. If you aren’t an expert in the subject, it can be hard to tell the difference until you find out the hard way.

————–

Subscribe
Notify of
guest

0 Comments