One of the mantras I often think about is a phrase uttered by data scientists, “Garbage in, garbage out”. The idea is that if you have bad data sources you will get bad results. This got me thinking on how ChatGPT and AI will work.
When musing on the future of how all this will change our lives, there is that quick instant of panic. That moment of realization that your entire purpose for existence can be replaced by this tool. Then I remember that using ChatGPT is a skill. All skills require a clear vision to build successful projects. Tools only go as far as the humans who drive them, for now… muahahaha.
Jokes aside, if you are afraid of, or bored by, something like ChatGPT, it may still be worth exploring it further. Today we’ll discuss things like:
Will AI take your job leaving you ranting like a forgotten South Park character?
Is this overblown hype and the next big fad?
Let’s talk about what I’ve learned about AI and ChatGPT in the last few days.
AI is not overblown nor is AI a fad
Remember when Siri came out and there were people who swore they’d never use voice command technology? Full disclosure, I was one of those people.
Then one day Spotify gave us free Google Home Mini’s. Over time terms like “Ok Google, what’s the weather?” and “Ok Google, set a timer for X minutes” became a regular staple of my life. I don’t use many voice commands all things considered, but I do use them and so do most of us.
It may take place when you are driving. Voice to text is simpler than typing on your phone while avoiding a ticket. Some people went full automated, and their Google powered homes are set up that you can adjust every possible house related setting with voice alone.
If voice command doesn’t clearly display an example of technology adoption that sells the longevity of AI, consider texting. The internet is full of memes of grandparents discovering the joys of SMS and modern chat. The point is, as time goes on, everyone inevitably gets on board. These game changing scary technologies become regular staples of life.
Your grandma may never use her phone with the finesse of a 19-year-old baddie on TikTok, but she still knows where the poop emoji button is. People will adopt tools like ChatGPT in different ways but we will all end up using them.
There are physical limits to modern day AI
There are clear limits in the ability of what artificial intelligence technology can accomplish in 2023. There are also limits in a physical boundaries way that AI companies are encountering. Apparently it takes a lot of space to create server rooms with the power to do the AI magic we all love or hate.
I came across this video:
Subscribe to Goda Go as her content is wonderful. For those skipping the video, it turns out Midjourney does not advertise on purpose. The computational power required to sustain their userbase is significantly higher than that of their social media counterparts.
This means it takes a lot of computers to support the current ecosystem. They have even spaced their systems around the world to take advantage of constant nighttime temperatures. Computers overheating is a big problem when you are doing this nature of supercomputer work. Nighttime use means less power consumed and less cost.
There is no clear path for them to scale or grow. More users mean that they need more physical data centers to process all the data. This puts a cap in how big any one company can grow right now. There will be Midjourney alternatives but if things like building complexes are the barrier, then for now the race to human obsoletion is stunted.
This is temporary. One day there will be new CPU technology that will allow people to get quadruple duple uple power with significantly less energy and space.
The secret to ChatGPT is good prompts
A lot of people may have boring first experiences with artificial intelligence. They open their Snapchat AI friend and ask it bad questions. I don’t mean the questions are inherently bad, only that they are poorly worded prompts. The question poorly worded for the goal of a good AI answer.
A prompt is the jargon term for when you ask the AI to do something. You are prompting a response by asking it to do something specifically. I’m still learning prompt etiquette but the specific order of words you use makes a difference. What’s cool about prompting is you can use prompt commands to prime ChatGPT to give you the answers you want.
By default, when you ask it for something, it answers generically. You are letting the language model loose into the wild and hoping it adopts an answer that fits. To increase your chances of success you can ask ChatGPT to roleplay for you.
If you ask it to answer a math question, it will answer it. To make sure you get the correct answer you can ask ChatGPT to pretend to be a brilliant mathematician. This makes sure it answers using the frame of reference “brilliant mathematician”. This matters because without the specific role, ChatGPT may decide to take on the role of “guy who got a C in math” to answer the question.
You need to engage in roleplay with the AI
To get the most out of ChatGPT you need to craft your prompts the way a playwright would craft a scene. You set all the context and then you bring in the characters and dialogue.
Within an individual ChatGPT chat thread, the AI will remember things based on the prompts you give it. If the first prompt is to ask ChatGPT to assume a specific role, say a hiring manager at the job you have an interview for, it will now take on that character for the remainder of the chat.
Using good prompts, which you should save, you can create characters for you to roleplay with. After defining its role, you can tell ChatGPT what your role in this scenario is.
Perhaps you are applying for a job. Try this:
Copy/paste the job description and tell ChatGPT it’s hiring for this role.
Provide it with your CV and experience.
Ask it to craft a cover letter & update your CV to fit that job description.
Ask it the main pain points this hiring manager would feel.
Ask it what interview questions this hiring manager would ask.
Etc.
You can choose a very conversational tone with Chat GPT. However, from what I understand you can’t have naughty time with ChatGPT, go find a dating sim if that is your goal.
Shot prompting will make your life easier
When you ask the AI something outright without additional context, it’s known as “0 shot prompting”. Basically, you ask the question directly. An example would be:
List 5 cool titles for the blog below following best SEO practices: (blog text).
This prompt will then list you 5 titles. However, ChatGPT has no frame of reference to know if these titles are good or not. The tool has not been connected to the internet proper since September 2021, it’s best practices may be out of date.
You can leverage a tactic called “shot prompting”, which primes your prompts with more information, to contextualize what a good result should look like.
An example would be:
List 5 cool titles for the blog below based on the best SEO practices. Here is what 5 successful blog titles look like for this blog: (List of 5 titles) Here is what 5 unsuccessful blog titles look like: (List of 5 titles)
Use this as a reference when analyzing this blog text for a title: (Blog text)
By giving it a simple example of what success/failure looks like, your results will be better. The art of ChatGPT is going to be the art of context. To take it even further, if you have actual data, such as the click-through rates of those titles you can train ChatGPT to understand your individual needs.
Once you craft mult-shot prompts, it’s a good idea to save them. You will be able to reuse them to get ChatGPT where you need it to be with a few copy/pastes.
ChatGPT (or some tool) will enhance your life
I told you dear reader that I would give you an update on how ChatGPT improved my job hunting I used its criticism mode to evaluate my CV and rewrite based on what would attract a recruiter. I put this version of my CV on Indeed and let it be.
Immediately I was getting more emails for job positions. Indeed, will basically send you emails saying that someone is interested.
Within days of posting this I got a callback from a recruiter and have now gone through 2 successful interviews. Without being too presumptuous, it is looking like if I pass the background check I will get an offer.
I could have spent hours researching how to redo my CV, instead ChatGPT did it instantly. Then Indeed, another automated tool, fed me a job. It’s really wild to think how far we’ve come to walking in with a stack of paper CVs and filling out applications.
It’s not just job hunting. I met this girl who was struggling with a deadline for her Ph. D thesis. I broke how she could use prompts to have ChatGPT organize a plan of attack to meet her deadline. ChatGPT can create a schedule that takes into account your regular life obligations.
There are so many real-life applications for you to use this tool to change your life, I promise. This all has one caveat.
Using AI to cheat is going to get you caught
Given how magical AI is, AI detection tools will be just as magical. Already they can scan text and other content and give you a breakdown of what % is human written and what % is AI written. That’s bad news for a lot of people. My assumption is you will have to paraphrase every word to get around detection.
At least if you keep “track changes” on while using MS Word you’ll be able to prove it’s you. I bring this up because if you use ChatGPT to write everything for you, you may face some ownership issues down the line.
I will ask ChatGPT to tell me if this blog article is clear. I will ask it to point out grammar mistakes. These are things I’d ask any editor. Using ChatGPT this way is legit AF. I wrote a text and am using an assistant to validate quality.
In the same way if you write code, ChatGPT can point out the sections that contain bugs or can be rewritten cleaner. If you are asking it to edit your code, you still own that code you wrote.
The bots people create entirely with ChatGPT will face some ownership issues when other people steal the idea. Once people realize your idea is just AI derived, it’s fair game in the land of IP theft. With great magical tools comes great understanding of the legal boundaries of ownership.
I am not a lawyer so consult with one for real business.
The ethics of AI and ChatGPT
I believe these tools should be used ethically. To me that means that they should not be used to cause harm to others. Unfortunately, we all know that scammers and other types are going to have a field day with ChatGPT.
Keep in mind that OpenAI (the company that made ChatGPT) will have to comply with the authorities. They will have access to everyone who used these tools and can follow up. This will be the case for most of the knockoff versions of ChatGPT that exist.
Make sure to be careful with how you use ChatGPT. I plan on using it as a virtual assistant to help make my life better. Using it to plagiarize, steal or cut corners is going to end up hurting you. Those of you that are reading this with real sincerity are not the hacker types that know how to not get caught. You are the type that is coming to my blog to learn more about where ChatGPT is at.
I follow copywrite law, as I am a creative artist. Be mindful that the content generated by tools like ChatGPT and Midjourney could be based on the theft of other people’s intellectual property. ChatGPT scanned a lot of Reddit posts without consent. The “creativity” it displays, while a distinct entity, isn’t harmless.
Make sure to stay on top of the law surrounding AI. Use it for good and you should be fine.
If you need help let me know
There is a new job position called prompt engineer that is being invented to deal with ChatGPT. Many people are going to encounter difficulty communicating with AI tools. As much as prompts will be a matter of copy/paste work, there is an art to this that will improve your output.
Garbage prompts will lead to garbage answers. Artfully crafted prompts will be life changers. Stringing prompts together can create macros (these strings of prompts that I believe you can own as the combination of prompts is your creation).
I’m going to keep learning how to use the tools and resources related to AI automation. As I learn them I will develop courses and content around them where you can learn. Some of it will be free, some of it will be gated.
We are heading in a direction where it will be, pay me and I’ll just do it for you, and you can skip the learning part all together. Basically, how it goes with all new technologies. To put it bluntly, pay me and I can teach you.
Last few days have been busy and I’m going to NYC on Sunday. I imagine the blog slows down to one every 3-4 days for the next couple of weeks as life adjusts. Until the next time.
Live Long and Prosper Everyone