Demystifying ChatGPT and Other AI Large Language Models

| 0

March 2025

Two years ago, I sat down in front of my computer for an unusual interview—not with a person, but with an AI large language model — ChatGPT. Back then, I was curious. Could this AI really hold a conversation? Did it understand what I was asking? And most importantly, where was it getting its information? The AI answered my questions smoothly—almost too smoothly. While I was impressed, I walked away with a nagging sense of mystery. I didn’t understand how it worked—just that it did. More importantly, I largely dismissed it as a useful tool due to highly publicized examples of how it had gotten things wrong and its tendency to hallucinate. Frankly, I was skeptical. (If you’re interested, you can find the original interview here.)

The blog turned out to be popular, far more so than many others I have written over the years. Much has happened in the last two years—not just with ChatGPT, but with many other large language models –or “LLMs”. Without question, LLMs have improved significantly since my “interview” with ChatGPT . They are faster, more accurate, better at citing sources, and in some cases able to access the internet for current data. 

LLMs have also have “gone mainstream,” becoming integral to our web searches on Meta, Google, and Bing. The same technology is now embedded in tools that offer “helpful” rewrites of our emails, texts, and documents, and it is a key component of most online and telephone help desk inquiries.

But how does an LLM actually work? Why is it so good at sounding human? And what’s really going on under the hood when it generates a response?

It’s time for an updated conversation—one that does a more complete job of explaining how these AI programs work, and what are the benefits and risks of incorporating them in our daily lives  

Tokens, Parameters, and the Magic of Predicting “What Comes Next”

One of the most common misconceptions about LLMs is that they “think” like we do. They don’t. There’s no “ghost in the machine,” no digital consciousness weighing options and forming opinions. Instead, it’s more accurate to think of LLMs as sophisticated probability engines that “guess” the next word based on a mathematical model programed and modified using vast amounts of text data.

To unpack the magic of an LLM, it helps to understand two key terms: tokens and parameters. Of course, there are many more—a cacophony of technical jargon thrown around by engineers—but these two will suffice for our purposes.

Tokens

Think of a token as a chunk of text. It could be a single word (like “and,” “to,” “from”), or more likely a syllable or part of a word (like “un-“, “dis-“, “-ly”). When you type a question into an LLM, it doesn’t “see” the words in the sentence as you do. It breaks them down into tokens (words and bits of words). This is critical because, in the next step, the model must analyze token relationships using probability weightings, to generate a response.

The latest versions of LLMs can handle over 100,000 tokens in a single session. That’s impressive, especially considering that a well-educated adult has a vocabulary of about 40,000 words. This expanded token limit allows for even more complex and extended conversations, enabling models to process and generate larger pieces of text in a single interaction.

A Trillion Parameters: The “Rules” Behind the Response

However, the real magic happens when the LLM selects the next word (or series of tokens) in response to our question or request. This is done using a set of mathematical probability weightings assigned to various token combinations that comprised the question or request you typed or ask. These weightings are called parameters.

If tokens are the building blocks—the “dictionary”—then parameters are the “rulebook” that governs how an LLM constructs coherent responses. Modern LLMs have a staggering number of parameters. For example, some of the most advanced models are estimated to have nearly 1.8 trillion parameters! To put that in perspective: if you looked at each parameter for just one second, it would take over 57,000 years to examine them all! 

“Training” the LLM

These parameters aren’t something an LLM invents on its own. They are shaped by human input— and analysis of word structures from massive datasets. These datasets incorporate a large percentage of human written knowledge and literature. Human feedback and carefully designed adjustments to parameters have been added to most LLM’s to keep responses accurate and safe. Over time, the responses generated by the LLM are refined, corrected, and guided by human oversight.

These parameters enable an LLM to generate understandable responses, that are almost always correct and accurate, because relative values assigned to the words used and their order, are based on vast amounts of analyzed data. 

This is one function of the massive data centers that have emerged worldwide, where models are trained and updated on enormous datasets containing text, and in some cases, audio and video data. By examining an extensive portion of human knowledge, the LLM produces outputs that mimic human speech. Over time, as these outputs are tested, refined, and corrected and updated through reinforcement and human feedback, they become more contextually appropriate and accurate as well.

For example, the LLM determines whether the word “bank” refers to a financial institution or the side of a river based on surrounding words and user context. It can even gauge the tone of a sentence—whether a phrase is humorous, serious, or sarcastic.

Here’s an example based on a real exchange with ChatGPT:

Me: What can cows jump over?

ChatGPT: Cows are not known for their jumping abilities, but they can jump if necessary. In general:

  • Cows can jump over low fences (typically up to 4-5 feet if motivated).
  • Younger, more agile cows may jump higher than older or heavier ones.
  • If startled or frightened, cows can clear obstacles they normally wouldn’t attempt.
  • The phrase “the cow jumped over the moon” from the nursery rhyme is, of course, just a fun exaggeration.

Me: Rhyme: What do cows jump over.

ChatGPT: Cows can jump, but not too high,
A fence, a log, if they really try.
Over a ditch, across some hay,
But not the moon—no, not today!

In the first response, the LLM assumed I wanted a factual answer. In the second, my input “rhyme” triggered a different set of parameters, generating a poetic and more light-hearted response.

Limitations: When AI Gets It Wrong

Understanding how LLMs work also explains why they sometimes get things very wrong. Since they rely on predicting statistically probable next words, they can hallucinate information—confidently stating incorrect facts because they “sound right” based on training data. This is why AI-generated content should always be double-checked, especially for accuracy-sensitive topics.

A striking example of this occurred when lawyers submitted a legal brief that was generated by an LLM. The brief cited multiple legal cases that simply did not exist. The lawyers, assuming the AI’s output was correct, did not verify the cases before submitting them to the court. When the judge reviewed the citations and discovered that none of them were real, the attorneys faced serious consequences, including fines and reputational damage. 

In addition, unless an LLM has real-time access to the internet and is updated regularly, its knowledge may be outdated. The old adage “garbage in, garbage out” applies. If an LLM is trained on flawed or biased data, its responses may reflect those inaccuracies. For example, if a model is trained on outdated medical research or biased historical accounts, its responses may reinforce those inaccuracies. 

Trust But Verify

So how can you verify the accuracy and reliability of information provided by an LLM?

  • Cross-check claims against reliable sources such as academic journals, news outlets, or expert databases.
  • Ask the LLM for its sources. While it doesn’t always provide direct citations, some models can generate references or indicate the general sources of their information, offering a starting point for verification.
  • Use models that cite sources. Some LLMs, like Perplexity AI, not only restate your query to ensure proper interpretation but also provide footnoted responses with references to specific sources, allowing for greater transparency and verification of information.

Take Time to Read and Consider AI Responses

One of the best habits to develop when using an LLM, is to slow down. Humans tend to trust authoritative responses, especially when delivered with certainty. But because they are a product of statistical probabilities rather than a conscious response, AI-generated answers should be carefully evaluated against your real-world knowledge and common sense. One method that can help in engaging critically, is to ask the LLM follow-up questions, request clarifications, and challenge responses. This will not only improve accuracy but it also ensures a deeper, more complete understanding of the topic.

A Powerful Tool

LLMs are powerful tools, but they remain just that—tools. They don’t replace human reasoning, intuition, or expertise, but they can significantly enhance them. By understanding how LLMs work, we can engage with them more effectively, apply them more appropriately, and develop a balanced perspective on their strengths, limitations, and best applications. Written with the assistance of ChatGPT and Perplexity AI

Online Games – The Benefits and How to Stay Safe

| 0

One of the most popular uses of the internet is online gaming. Smart phones, gaming consoles, computers, tablets, and virtual reality devices all allow us to participate in online games alone or with others.

While gaming may seem like a waste of time to some, there are potential benefits to online gaming for everyone: 

  • Games can help develop cognitive skills through strategic thinking and problem solving.
  • First person games (where you have the point of view of your character in the game) can help improve hand-eye coordination and fine motor skills through precise movements. Others help enhance cognitive abilities like memory, attention, and spatial awareness.
  • Multi-player games provide an opportunity to connect with friends and family as well as meet new people from around the world. Online communities develop and thrive through games and help give people a sense of community, especially for those that find it difficult to socialize in person. Team-based games also help to promote teamwork, leadership, collaboration, and communication skills.
  • Gaming can provide stress relief, an escape from everyday pressures, and provide relaxation and help improve mood.
  • Gaming helps teach overcoming challenges, dealing with setbacks, and multitasking skills as gamers learn to adapt new strategies and keep trying until they succeed.
  • Some games incorporate historical settings, scientific concepts, and language to make learning interactive and fun.

While gaming can provide all these potential benefits, there are potential threats to be mindful of while enjoying online games. The threats gamers may encounter include:

  • Hacking of game accounts to steal valuable in-game items, currency, and personal information.
  • Phishing scams via email, websites, or in-game scams that trick gamers into revealing personal information, login credentials, and payment details.
  • Malware, trojans, and viruses that are often disguised as game modifications\customizations or cheat codes leading to infected systems and stolen data.
  • Cyberbullying and toxic behavior during and after game play that affect players’ mental health.
  • Deepfakes and voice cloning can lead to identity theft, misinformation, and harassment.

How can we protect ourselves when gaming online?

  • Protect your personal information. Use two factor authentication to protect your accounts along with strong password protection. Make sure that you also set up ways to recover your accounts should you forget your password, or your account is hacked. When setting up accounts, only provide the minimum personal information needed to set up your account. If games have a subscription, use a third-party payment system such as Apple Pay, Google Pay, PayPal, etc. This keeps credit card information out of your account information.
  • Parents should set up parental controls for children, making sure they have access to age-appropriate games. Set time limits on playing as well as when games can be played where possible to make sure kids spend time with friends and family outside of games. Teach your kids about online safety and that they should talk to you or another trusted adult if they feel threatened or bullied when gaming online. Help educate your kids about potential scams online and in games, and make sure they talk to you before purchasing or downloading games or in app purchases.
  • Make sure PCs, tablets, and gaming consoles are updated to the latest operating systems and gaming software. PCs and tablets should have an up-to-date antivirus program on them as well as gaming consoles where possible.
  • As gamers and parents, we should also make sure that we stay safe mentally as we game. We need to protect ourselves and our kids from cyberbullying by playing games with friends and family and only participate in gaming communities that promote positive and healthy gaming habits. Players that bully or create a toxic environment for you and other players need to be reported to the people or companies hosting the games and blocked where possible.

Gaming addiction is real, and both adults and youth need to set up ways to help reduce the possibility of gaming addiction, identify when it happens, and seek help if they become addicted to online gaming.

Avoid gaming addiction by doing these practices:

  • Set specific time slots to play games during the day. Use alarms to remind you when to take breaks from the game and engage in other activities, rest your eyes, stretch, or just take a walk.
  • Make sure to prioritize responsibilities such as work, school, chores, sports, and time with family and friends before gaming. Use gaming as a reward when specific tasks are completed.
  • To help reduce eye strain, try to follow the 20-20-20 rule. Every 20 minutes of screen time or game play, look at something 20 feet away for at least 20 seconds.
  • Make sure to engage in other activities. Balance gaming with other hobbies such as reading, sports, or creative projects. Make sure to spend time with friends and family face to face.
  • Monitor gaming habits. Keep a log of how much time is spent gaming and how that time impacts daily life. Adjust the time spent gaming to make sure you take care of yourself, and your other responsibilities. Also monitor in-game purchases and set a budget for gaming expenses.
  • Create game-free zones in the house such as the dining room or bedroom as well as non-gaming hours. Try to keep the devices you use to game out of site or at least off during non-gaming hours.

If you or someone you know is addicted to gaming, seek support from friends, family or find a professional that you or your loved one can talk to. There are support groups available to help people through gaming addiction. Check sites such as Gamming Addicts Anonymous  to see if there is a local group near you.  They also provide online meetings where local groups are unavailable.

Online gaming provides wonderful opportunities not just for fun but for learning and connecting with others. By balancing the joys of gaming with safety measures, you can ensure that online gaming experiences are both fun and secure.

For more information:

Securing Online Gaming:

Cyberbullying:

Gaming Addiction

Cyberbullying and How to Deal with It

| 0

In 2022 19.2% of students surveyed nationwide, in grades 6-12 (ages 12-18), who reported being bullied during school, about 21.6% reported being bullied online or by text, including nearly twice as many girls (27.7%) as boys (14.1%) (National Victim Survey – School Crime Supplement).

One of the many reasons people may not want to connect online or use online applications is cyberbullying. Cyberbullying can have long-term consequences for the mental health of both youth and adults. It is important to understand how to prevent and deal with cyberbullying, should it occur to you or someone you know.

What is cyberbullying?

  • Cyberbullying takes place over digital devices like cell phones, computers, and tablets. Cyberbullying can occur via texting and messaging apps, or online in social media, forums, or gaming where people can view, participate in, or share content.
  • Cyberbullying includes sending, posting, or sharing negative, harmful, false, or mean content about someone else. It also includes sharing personal or private information about someone else to cause embarrassment or humiliation.
  • Cyberbullying can cross the line into unlawful or criminal behavior.  Each state has different laws regarding cyberbullying.  You can learn more about the laws regarding cyberbullying in Missouri through the Cyberbullying Research Center  

How does cyberbullying occur?

It’s important to understand how cyberbullying occurs so it can be easily recognized, and we can take action to help children deal with cyberbullying. Some of the common tactics are:

  • Posting comments or rumors about someone online that are mean, hurtful, or embarrassing.
  • Threatening to hurt someone or telling them to kill themselves.
  • Posting a mean or hurtful picture or video.
  • Pretending to be someone else online in order to solicit or post personal or false information about someone else.
  • Posting mean or hateful names, comments, or content about any race, religion, ethnicity, or other personal characteristics online.
  • Creating a mean or hurtful webpage about someone.
  • Doxing, an abbreviated form of “documents”, is a type of online harassment used to exact revenge and to threaten and destroy the privacy of individuals by making their personal information public, including addresses, social security number, credit card and phone numbers, links to social media accounts, and other private data.

How can you tell if someone is possibly involved in cyberbullying?

The first step in preventing cyberbullying is to be aware of what kids are doing online. You need to know if your child is a victim of cyberbullying, witnessing cyberbullying, or is bullying other children online. Warning signs you should look for as a child uses their device include: (From stopbulling.gov)

  • Noticeable increases or decreases in device use, including texting.
  • Exhibiting emotional responses (laughter, anger, upset) to what is happening on their device.
  • Hiding their screen or device when others are near and avoids discussion about what they are doing on their device.
  • Social media accounts are shut down or new ones appear.
  • Avoiding social situations, even those that were enjoyed in the past.
  • Becoming withdrawn, depressed, or loses interest in people, hobbies, and activities.
  • Complaining about illness or physical symptoms to get out of going to school.
  • Difficulty sleeping. Almost 63% of victims report a lack of sleep as a psychological impact of being cyberbullied.
  • Talking about self-harm.

If you suspect your child may be involved in cyberbullying, take steps to investigate their digital behavior. Take the same approach as you would to any form of bullying. Make sure to provide support to the child being bullied, address the behavior of the bully or bullies involved, and make sure everyone understands that cyberbullying is taken seriously and can have serious consequences.

How can parents and youth prevent cyberbullying?

Responding to cyberbullying does require different approaches to address the issue. Some the things you can do to address the issue are:

  • Notice- Recognize if there has been a change in mood or behavior. Try to determine if these changes happen around a child’s use of their digital devices.
  • Talk – Ask questions to learn what is happening, how it started, and who is involved. Use active listening skills to get below the surface of conversations with youth.
  • Document – Keep a record of what is happening and where. Take screenshots of harmful posts or content. Most laws and policies note that bullying is repeated behavior, so records help to document it.
  • Report – Most social media platforms and schools have clear policies and reporting processes. If a classmate is cyberbullying, report it to the school. You can also contact app or social media platforms to report offensive content and have it removed. If a child has received physical threats, or if a potential crime or illegal behavior is occurring, report it to the police.
  • Support – Peers, mentors, and trusted adults can sometimes intervene publicly to positively influence a situation where negative or hurtful content posts about a child. Public intervention can include:
    • Posting positive comments about the person targeted with bullying to try to shift the conversation in a positive direction.
    • Reaching out to the child who is bullying and the target of the bullying to express your concern.
    • Rallying around a victim of cyberbullying is a highly effective defense against bullies. Bullying works by trying to isolate a victim or have them feel isolated. When friends and other youth rally around a victim, it can thwart the bullies.
    • Determining if more professional support is needed for those involved, such as speaking with a guidance counselor or mental health professional.

  • Teach internet safety. Making sure they understand about privacy, predators, phishing schemes, catfishing, and security.
  • Learn about their online friends and what they are doing online.
  • Let them know that you want to be aware of their online life and you will want to check on their online activities, friends, and files.
  • Help youth manage cyberbullying. Discuss what cyberbullying is and make sure they take any threats seriously. Teach them how to react to potential cyberbullying by ignoring or responding assertively and calmly to the person or persons doing the bullying. Identifying trusted adults that the child can speak to about potential cyberbullying.
  • Help youth build and maintain self-confidence. Having healthy self-esteem allows youth to be able to ignore or address cyberbullying.
  • Make sure youth understand you don’t approve of cyberbullying in any form and hold them accountable if they participate or witness cyberbullying.

What can youth do to help prevent cyberbullying?

  • Never respond to harassing or rude comments.
  • Save or print the evidence.
  • Talk to parents or a guardian if harassed; get help reporting this to your internet service provider, school, or local law enforcement.
  • Respect others online.
  • Only share passwords with a parent or guardian.
  • Change passwords often.
  • Password protect your cell phone.
  • Use privacy settings to block unwanted messages.
  • Think before posting or sending photos – they could be used to hurt you or someone else.
  • Contact the site administrator if someone creates a social networking page in your child’s name.

For more information on Cyberbullying:

What is Cyberbullyinghttps://www.stopbullying.gov/cyberbullying/what-is-it

School Crime Supplement– National Center for Education Statistics – https://nces.ed.gov/pubs2024/2024109rev.pdf

What is cyberbullying and how can it be prevented? – Mayo Clinic – https://mcpress.mayoclinic.org/parenting/what-is-cyberbullying-and-how-it-can-be-prevented/

Bullying – Mo Dept of Elementary and Secondary Education – https://dese.mo.gov/governmental-affairs/public-school-laws-missouri/bullying

Courage2Report – MO State Highway Patrol- https://www.mshp.dps.missouri.gov/MSHPWeb/Courage2ReportMO/index.html

Cyber-bullying: What Parents Can Do About it – The Center for Parenting Education –

Cyberbullying: Learn the Signs – American Society for the Positive Care of Children – https://americanspcc.org/cyberbullying-learn-the-signs/

How to help Kids Deal with Cyberbullying – Child Mind Institute – https://childmind.org/article/help-kids-deal-cyberbullying/

Mizzou Ed Bully Prevention Labhttps://www.mizzoubullypreventionlab.com/