Disinformation and misinformation work for a very simple reason. People who did not study history, and do not follow current affairs do not have the background information to spot what is true from what is untrue. They are unable to use history’s lessons to see the warning signs and act accordingly.

Algorithms and Social Media

In Mindf*cuk by Chris Wylie we read about how big data and algorithms were used to identify and exploit people’s gullibility to get them to vote for Brexit in England, and the Republican candidate in the US. Algorithms, for years have controlled what we see, by whom, and how often. Facebook has been caught more than once playing with algorithms to intentionally make people miserable as well as to manipulate people in to seeing things one way, rather than the other.

Yesterday chatGPT was down, yet again, so I browsed through the web looking for other Large Language Models to experiment with. In the process I came across personal AI. Personal AI is an app, at the moment, that runs on your laptop, but will eventually come to mobile phones and more.

The premise is simple. You tell the AI things about you, and it then returns answers based on what you told it. You can chat with it and it will give a response based on what you tell it.

Playing with AI and Learning to Keep It Simple, Silly

Yesterday I was playing with AI in the evening and I asked it to help me write a function that would detect whether an array item was a photo or a video. I told the AI, this is a photo array item, and this is a video array item. I want the photo array item to display the img src code and for the video item to be displayed with video src code. At first it failed to detect anything and gave me an error message. I decided to highlight the photo array item snippet and the video array snippet. It didn’t come back with an error but it did not display any img or video code. In the end I decided to say “here’s the img src tag. I want you to detect whether the file is mp4 or jpg. If it’s mp4 add the video player, and if it’s a photo load the image. I went from asking it to read the array to determine what to do, to “check the filetype and load the appropriate src type. That worked, and required very little code. Don’t Over-Engineer With AI The moral I came away with, after the trial and error I went through last night is that whilst AI is great at reading and understanding code, and giving us answers it can sometimes fail. I was making something far more complicated than it had to be and chatGPT got confused, as would a human trying to complicate something simple. PHP has a lot of functionality built into it. It has tools that allow it to read exif information, file types and more, and if you know about these tools, then you can come up with elegant solutions, rather than complicated ones.

Today I want to discuss something different. I want to discuss using AI as a teacher, rather than as a tool to replace us. For a few days I have been playing with the Instagram json file that I got when I requested my archive. With this json file I thought about things to do, and in so doing asked chatGPT for help, and it provided answers.

Adapting the Questions

They love to say “Don’t fear AI”. They all say “AI is meant to help people with their work”, “AI is meant to be a personal assistant” and more. It’s not AI we fear. It is stakeholders, governing boards and accountants. They are the people we fear. Throw in Inhuman resources for good measure. We “fear” AI because they, all the groups mentioned above, want to use AI as an excuse to fire human beings because computers are theoretically, and only theoretically, cheaper. 

If you follow chatGPT and AI courses you will often come across the term “Prompt Engineering”. Prompt Engineering is the art and science of writing instructions so that AI understands what you would like to do. With time and practice you can be as specific as you like. You can fine tune your request until you get precisely what you are looking for. 

**The Limit of chatGPT. **

“As an AI language model, my responses are generated based on a mixture of licensed data, data created by human trainers, and publicly available data. These sources may contain publicly available information up until my knowledge cutoff in September 2021. It’s important to note that I do not have direct knowledge of specific data gathering practices or the individual sources used to train me.”

I am old enough to remember a teacher writing on a board or piece of plastic for an overhead projector. “Why don’t you just give us photocopies of what you’re writing instead of asking us to copy down what you’re writing. “Because you will remember it better if you write it down.”

At the time this seemed stupid and a waste of time. Years later I think that we could have been taught to take summarised notes rather than literal notes but that isn’t the point. The point is that learning is as much about writing as it is about understanding the material Every day for weeks in a row I have taken the time to write a 300 word post and it’s difficult. Not only do you need to come up with an idea but you need to develop it into something that is at least three hundred words long.

Time for a new discussion to take place. Reading time. Do you read through articles or do you skim them. Is reading the headline enough or do you read every word of the article? I ask because in the age of chatGPT and GPT3 I would ask the same question as I asked about social media.

If you want to discuss ROI for businesses and PR firms or advertisers then you need to discuss ROI for users too. Facebook, Instagram and Twitter wanted to control the timeline by using a “home” feed. The Home feed is an algorithm driven feed that shows you popular content, rather than content by friends, that you and they would converse about. The result is a lonely social media experience. The ROI goes down, That’s why I took a break from Twitter to play with Mastodon and revive my blogging habit.