HSB - AI Partnership

What are Generative AI & Large Language Models (LLMs)?

AI & Generative AI

Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks that typically require a human operator. Now imagine a computer program that can be creative! Generative AI is like that - it uses its knowledge to make new things, like text, pictures, or even music.

LLMs

Large Language Models (LLMs) are a kind of generative AI that focus on language. Because they've been trained on so much data, LLMs can produce human-quality writing, and even mimic different writing styles. They're also excellent at translation and working with data in multiple languages.

AI uses many techniques to simulate a human's cognitive abilities

Uses algorithms and statistics to analyze historical (training) data and become more and more accurate at predicting outcomes.

A subset of machine learning that arranges algorithms to mimic the human brain (artificial neural network).

Started over 50 years ago in the field of linguistics, it gives computers the ability to understand human language. It can be spoken or written in nearly any language.

Uses machine learning to identify and understand objects and people in both images and videos.

AI that learns by trial and error in a special environment. It gets rewarded for good choices and penalized for bad ones. This way, it refines its actions to maximize rewards.

What AI Is Not

Sentient or Conscious

While the current large language models like ChatGPT have very good conversational style, it is important to remember that they are trained on enormous amounts of human conversation and are designed to predict what you would like to hear.

Emotion, Understanding, & Subjective Experience

AI is simply complex software running on powerful computers. It is quite effective at simulating emotion and understanding, but it is merely because of the vast training data consumed. At the end of the day, it will lack the subjective experience that a human has. It cannot experience real-world sensory stimulus, and process the emotional & cognitive impact.

Infallible or Immune to Error

AI can produce its own "facts" and fake experience. In this field, they are referred to as hallucinations, but essentially it means the large language model is lying to you. While everyday the organizations producing the models make progress to reduce these, they are still hard to prevent and filter out. The end goal of most of these models is to tell you what you want to hear.

Where AI Falls Short

IP Infringement

AI works because of large amounts of training data. Most of the big firms building AI, have simply used the vast sums of information available on the Internet for their training. This brings into question who actually owns the “works” created by the AI.

Lack of Equity

Big tech names like Google, Microsoft, and others have released free AI tools. The concerning thing is that one company, OpenAI, is leaps and bounds ahead, and their product is paid only access. Going forward, funding models will determine if there is a gap in access.

Full of Bias

When designing a prompt (asking a question) for the AI, we’re inputting our own bias. Once again, these large language models aim to please, so it crafts an answer target specifically to us. The adage “garbage in, garbage out” is accurate here, and a huge problem that hasn’t been solved.

High Costs

This issue is multifaceted. In the first sense, if you’re using these tools, you are the product. You’re providing free training data with every prompt you create, and if you use the free tools from big tech companies, all your provided data is used for targeted advertising. On a deeper level, if you’re the company training these models or leveraging these models, the computing costs are astronomical.

AI May Provide Opportunities

Brainstorming

ChatGPT shouldn't be used to replace creativity, but you can use it to get new ideas and perspectives that should allow you to brainstorm.

Editing

In classes that are not grammar and composition related, students can use large language models to help clean up their work. This can already be achieved already in MS Word, Google Docs, and with a third party plugin like Grammarly.

Time Saving

Why not let students speed up their workflow? Allow them to use Generative AI to create document outlines, provide structure to a process, and even rough drafts for certain applications.

New to AI? We're here to help!