10 AI terms you should know 🤓
and how to use them to crush your goals
ChatGPT, Bard, Prompting, Machine Learning, AI are all terms that are now household words heard and spoken by kids and adults at the dinner table - they are as common as Google and Zoom.
Knowing these terms will get you far enough in your AI journey but there are several others, like Anchoring and Temperature, that you should be familiar with.
Let’s take a look at 10 additional AI terms you should know and why you should care. ✨
🎯 10 AI terms you should know
1- Anchoring ⚓
Anchoring refers to our cognitive bias, which gives us the tendency to rely too heavily on the first piece of information we get when making a decision.
What does this mean for using AI tools?
Your first prompt is the most critical. It will set the stage for the rest of your conversation with the AI. Within a prompt, you can also anchor the AI by giving it keywords early on that you want it to focus on. This is true for ChatGPT, Bard, and even MidJourney.
2- Prompt Injection 💉
Prompt injection is the act of trying to squeeze a prompt within a prompt and trick the AI into doing something it should not be doing. For developers of large language models (like those powering ChatGPT and Bard), this is a huge issue and something they focus a great amount of time on solving to make sure their models are not used maliciously.
For users of AIs, the issue is smaller scale but still present. If you write a prompt that is too complex and not orderly, you risk inadvertently injecting a prompt within your prompt, which will make the AI give you an answer that is not related to what you are looking for.
PS: read our article on how to use context with AI for help with this.
3- Bias 🧭
We are all now aware that social media tends to show you more of what you want to see and what the people running those platforms allow you to see. This is called the filter bubble. Unfortunately, AI has similar issues in that the developers of the AI can impact the bias the AI has by deciding what to train it on, which means that the answers it provides can be very biased based on those developers’ beliefs.
This is an issue that is worrying politicians, business leaders, and parents! When you (your team, or kids) are working with AI, keep in mind that the information you receive may be biased based on someone else’s views. Always double-check things!
4- Interpretability 🔍
AI models are often black boxes, which means that it is difficult to understand how they make decisions. This can make it difficult to trust the output of the model, and it can also make it difficult to debug the model if it is not performing as expected. Heck, even the developers of the models can’t necessarily know if the model is doing what it’s supposed to do, which is why you will often hear developers of large language models surprised about what outputs they get and how their models evolve with usage.
When you use a new AI tool keep that in mind and know that the AI could change with time.
5- Hallucinations 🦄
AI models are great at identifying patterns and then completing the pattern. Unfortunately, they sometimes tend to do this with information and facts as well, which means that when you get a response from an AI about something you asked the AI to look up or recall, you always have to double-check that it is true. It’s like that friend you had back in college that exaggerated every story they told.
An easy fix for this is when you get what seems to be fact by the AI, ask the AI to provide a source, and also ask it to confirm how accurate the information is. The AI is known to hallucinate but as far as we know so far it does not intentionally lie... yet?!
6- Model Limitations ⛔
The term is self-explanatory but this is a crucial point to know when working with an AI model or tool. If you are using a tool, try to find out what model they are using in the backend and understand the limits of that model.
The most common ChatGPT limitation we are now all aware of is that ChatGPT’s training was cut off in September 2021.
7- Feedback Loops 🔁
This is a concept where the AI's output is fed back as input for the next prompt. In chat models, like ChatGPT and Bard, this is often how a continuous conversation is maintained.
Just remember that when using Bard, you can’t come back to old conversations (as of this writing). You can only do that with ChatGPT.
8- Temperature 🌡️
Temperature is basically a setting that controls how much creativity the model should have (or how much it should be allowed to hallucinate). As of this writing, you can’t change the temperature of ChatGPT or Bard directly, although if you use the ChatGPT API you can change the temperature.
I am fairly sure this setting will be introduced to both AI tools at some point.
9- Fine-tuning 🔧
This is a process where an AI model is further trained on a specific dataset after its initial broad training. It's used to adapt the model to specific tasks or to make it do a certain behavior.
There are various new tools coming out that use GPT-4 in the backend but for specific use cases. These startups are most likely fine-tuning the overall model to fit their specific use cases.
10- Transfer Learning 🔄
This is a popular approach in AI where a pre-trained model is used as a starting point for a similar task. This approach saves time and computational resources as the model doesn't have to learn everything from scratch. For example, GPT-4 used GPT-3.5 and GPT-3 before that to get better and better. The training didn’t start over each time.
Hope you found these helpful! When exploring the fast-moving world of AI, keep these in mind to help you better assess which AI tools to use and also understand them better. 🎉
As always, if you have any questions, hit reply!
🛠️ AI tools to crush your goals
🔥 Feature your product on GPT Hacks
GPT Hacks is a fast-growing newsletter with over 1,500+ startup founders, business owners, and tech-savvy pros looking for new ways to leverage AI to crush their goals. Click on the button below to learn more.
⏮️ Catch up with articles from our archives:
How was today’s tip?
Rate today’s tip to help us make GPT Hacks more useful for you 🚀