Been hearing about AI a lot, because of the release of ChatGPT. Focus on natural language processing, artificial neural networks, and large language models. AI is more than that though. A lot of this has been behind the scenes for some time. Don't know when it turns into "AI."
Natural language processing provides AI with an understanding of the meaning and intent of content. AI doesn't know about anything; it knows how we talk about it. NLP systems have to content with English. English doesn't stick to its own standards.
Think of your content as a basket of bread, each loaf is a sentence, each word is a slice. In NLP it is called a "token." NLP can figure out meaning by breaking sentence into token and "somehow" understanding. Really learning patterns of speech, patterns that it is taught, as well as content we give it.
Artificial neural networks are under the covers. It's the framework for AI computing. Subset of machine learning. Inspired by the neural network of the human brain. Part of AI that's doing all the processing. Content creators don't touch that part. ANNs simulate a human brain. ANNs contain nodes, layers, weight, and bias. This is language-specific. A lot of bias in our content, some of which we are aware of and some of which we are not.
Machines don't understand truth. They can recognize things and predict things.
Generative AI and Large Language Models. Generative AI is a branch of AI focused on creating new information. Similar to examples. Looking at patterns. LLM subset of generative AI, focused solely on generating text. A system that knows what it knows, not continually learning, used NLU (natural language understanding), understanding more about language than out computers used to. Produces human-like text, understands jargon, idioms, sentence fragments.
LLMs use ANNs to process and create contextually accurate human0like text. Don't know what they are talking about, just know how we talk about it. LLMs, do not "know" anything in the same way we do. Discerns patterns in the language that we use to discuss a topic or concept.
Need well-structured data (content) to train AI well. The more consistent the body of content, the better it will be able to give good answers. If you don't train it, the deep learning sill still happen, but the results will be inaccurate, hallucinations, downright dangerous.
Clean your content. You must clean your corpus. If you do not, your AI will be unreliable. AI does not know if your content is accurate. If you have conflicting content, AI can hallucinate. AI makes stuff up. Beautifully written, grammatically correct.
No comments:
Post a Comment