Have you dabbled with Gemini to touch up an image? Have you chatted with Claude to automate screening clients? Maybe ChatGPT was useful for you when writing your profile or ad copy. As much as we might loathe Artificial Intelligence (AI), it can be useful and it certainly isn’t going away any time soon, so it’s worth knowing what’s going on in the background when you use it.
We know AI is just a computer in a data centre sending words and images back to you. But how does that actually happen? Chatbots aren't black boxes that conjure up things at random. They are systems designed, built, and maintained by humans.
There is a massive amount of hype around AI, but when we understand what it actually is—especially the role of large language models (aka LLMs)—we can learn its limitations. This helps us use the technology safely and with far less anxiety.
Understanding this matters because many sex workers are already using AI tools for advertising, administration, research, and screening. Because sex work has uniquely high risks, outsourcing critical decisions about your personal safety to a bot can have unforeseen consequences.
Knowing that these systems generate likely answers rather than true answers can help you determine when they’re useful and when they need a second look.
Guessing isn’t knowing
When you type a message into a chatbot, it isn’t “understanding” language in the way a human does. It doesn’t consider your feelings, form opinions, or draw on lived experience. It’s relying on patterns it has seen before to determine what to do next.
That’s why AI companies are so desperate for things to train their AI on. The more examples they can feed into a model, the more patterns it can learn, and the better it becomes at predicting what a user might want to see.
Take screening a client, for example. You’ve seen hundreds of text messages from clients and you’ve noticed that certain phrasings tend to come from time-wasters and other phrases tend to come from genuine enquiries. All your life experiences give you a gut feeling about which messages are worth responding to and which are going to be a dead end.
A chatbot does something similar, but with statistics instead of experience or gut feeling. It’s been trained on a massive amount of text and uses probability to guess what words it should display as a response to your prompt.
Finding the most probable answer
Before the chatbot can do anything with your prompt it chops your words up into pieces called “tokens.” A token isn’t always a whole word—it might be a few letters, punctuation, or part of a word. Those tokens are then turned into numbers.
The AI system performs vast amounts of mathematical calculations on those numbers. It’s not searching through a database of things it read during training. Instead, it’s using the patterns it learned during training to calculate what token is most likely to come next.
The result is a set of probabilities for every possible next token. What you see presented back to you is actually what the system predicts is the most likely response based on the patterns it learned during training. Not necessarily the correct answer, or even the truest answer, just the answer that appears most probable according to a mathematical formula.
Confident hallucinations
Because the chatbot is predicting what comes next rather than checking whether something is true, it can be wrong, and often it’ll be wrong with complete confidence. It might invent court cases that never happened, make up academic papers that don’t exist, or confidently provide medical information that simply isn’t true.
This is called “a hallucination.” A chatbot isn’t lying in the way a person lies—it doesn’t know whether what it’s saying is true or false. It’s simply producing words that statistically tend to follow one another. This is why AI can be incredibly useful sometimes, and then completely unreliable at others. Even its best probability calculations are still a guess.
It's not magic, it's just math
AI can seem mysterious, especially when it’s presented to us as intelligent, creative, or even superhuman. In reality, it’s a tool that uses data, mathematics, and probability to generate responses that look convincing.
That doesn’t make AI useless. AI can save time, help with writing, assist with research, and automate repetitive tasks. Some sex workers have found practical ways to use these tools in their day-to-day work, but it’s worth remembering that AI doesn’t replace experience, judgement, or critical thinking.
The more we understand how these systems work, the easier it becomes to recognise both their strengths and their limitations, and to decide when their suggestions are useful, and when they’re best ignored.
Got a tech question for Ada? She wants to hear from you!
Ada answers all your questions about tech, the online world, and staying safe in it. No question is too silly, no hypothetical is too far-fetched! Learn to leverage devices, systems, and platforms to your benefit.
){kind=link}
