One strategy that I have found helpful when prompting LLMs is refinement. Oftentimes, LLMs do not quite hit the mark on the first try. Refinement helps make the outputs more centered on what you are actually looking for. During the activity on Tuesday, I had Claude create a week-long workout plan for a college swimmer at the D-III level, which focused on the 200 Fly and the 200 IM, the two events I swim. I found that the main sets were far more physically demanding than a typical practice. I then brought these concerns to Claude, who asked questions about how the schedule should be restructured. The revised schedule (picture below) that Claude gave was better, but still had sets that looked correct on the surface, but would be exhausting and unlikely to lead to improvements if actually implemented. More refinement could help Claude develop a workout plan I could actually use, a strategy recommended by UT in Tuesday’s reading, saying that refinement turns prompt sessions into a lengthy “dialogue” with your chatbot.

Prompt 1: Give me a swim workout

Prompt 2: I am a college-level swimmer at the Division III level. Create another workout centered on training for the 200 IM.

Prompt 3: Create a training block for a week. My main events are 200 Fly, 400 IM, and 200 IM

Prompt 4: I have concerns that doing that many 200 flys and 400 IMs in a row will be highly taxing on my body

Claude then asked me this (in a pop-out menu with responses I could click):
Q: How would you like to address the load on Saturday?
A: Keep Saturday but reduce volume/intensity
Q: Which event feels most taxing on your body right now?
A: 400 IM

The most helpful prompting strategy I learned is being specific and clear about what I want the AI to do. Instead of asking something basic like “summarize this,” I learned to give more detail, like asking for a summary in my own words, with examples, or in a certain tone. When I do that, the response is way better and actually sounds like something I could say.

One prompt I used was: “Explain this concept in my own words, like I’m talking to someone outside of class.” This helped me understand the topic more because the response was simple and easier to relate to. It didn’t sound too complicated or like a textbook, which made it easier for me to remember.

This connects to what we talked about in class about how prompts shape the response. The better and more detailed your prompt is, the better the AI understands what you want. It’s kind of like giving directions—if you’re vague, you won’t get the result you want, but if you’re specific, you get something way more useful.

Louis, Jordan, Dame

At the beginning, the starter prompt sort of seemed useful, but I knew that the response would be vague. I asked ChatGPT to summarize Toy Story and then I asked it to “summarize Shrek.” Both of these summaries were 8-10 sentences, not nearly enough to fully grab details from a film. The prompt pattern that made the biggest difference was giving it structure. I asked for it to give me a paragraph for each scene and it was much more detailed and searched the web before it did it, so it clearly found the things from the internet. After this, we asked it to help us all run faster in a month and it spit out a plan without knowing any details about us. I then used the reverse conversation method and asked it to ask us details about our height, weight, diets, sleep patterns and other details and once we answered those, it gave us a more detailed, precise plan that would last a month and it gave us an end goal. Overall, I learned that AI is good for outlines and specific plans, but only with specific details. In order to get good results, the user must be specific with their prompts and their results will continue to improve, especially when involving health desires. This draws back to the prompt literacy text that we went over for the week of March 31, where we discussed the benefits and issues behind AI.

 

My prompt to ChatGPT was, “Make me a workout to help me get faster.” In response, Chat said,

“🔥 Day 1 – Acceleration (Start Speed)

Focus: getting fast in the first 10–20 yards

Warm-up (always do this)

• High knees × 20 yards
• Butt kicks × 20 yards
• A-skips × 20 yards
• 2–3 light sprints

Workout

• 10-yard sprints × 6 (full effort)
• 20-yard sprints × 4
• Hill sprints (or resisted) × 4

Rest: 1–2 minutes between reps (FULL recovery)

---

⚡ Day 2 – Strength + Power (Gym)

Focus: explosive legs = more speed

• Squats: 4×5
• Deadlifts: 3×5
• Box jumps: 4×3
• Walking lunges: 3×8 each leg
• Core (planks): 3×30–60 sec

👉 Lift heavy but controlled — power matters more than max weight.

---

💨 Day 3 – Top Speed & Form

Focus: running fast once you’re already moving

Drills

• A-skips × 2
• B-skips × 2
• Stride-outs × 2

Workout

• Flying 20s (build up, then sprint 20 yards) × 5
• 40-yard sprints × 3

👉 Focus on:

• Relaxed upper body
• Big stride, fast arms
• Don’t tense up

---

🧠 Day 4 (Optional) – Agility & Change of Direction

Short burst sprints (5–10 yards)

Ladder drills (quick feet)

Cone drills (5-10-5 shuttle) × 4″

This workout is okay, but it does not give me any ideas that appeal to my own personal needs.

This prompt would be helpful for a new athlete, looking to learn basic workouts to get faster.

LLMs may not give you the outputs you desire when your prompts are not specific and clear.

I took a digital content creation class last semester and learned how to use AI a bit more effectively and efficiently while avoiding bias. The best way to use AI in a better prompting format is to be specific while using connections so the machine understands it better. For example, in our readings, it is frequently mentioned how bias impacts results. On the internet, there are many biased cites and sources full of news that is inaccurate, and AI platforms gain information from these biased cites. This means that nearly every form of information a user finds could be inaccurate, which is why it’s very important to be proper when wording questions and statements to the machine.

After working in groups we used Ai Gemini, a more enhanced version of AI that is most similar to generate greater portions of information distributed. The people in my group were from Ohio, so we decided to ask about the brown’s history as a football franchise. We found that AI also had bias in the information it provided to us. For example, it said that the browns had the most championships, when they did not because the NFL was formed after their main franchise was established. During these findings we discovered that this AI platform was getting information from bleacher report which is often very reliable and made it more confusing. I later looked at the source, and it had correct factual information, it was just incorrect with how Gemini portrayed it.

This is proof that AI frequently hallucinates, and it isn’t completely accurate all the time. We discussed this in class on Tuesday with our readings on how AI hallucinates and why it does it. This can come from things like bias, misinformation, and inaccurate data. It is important that we noted how AI is bad with numbers as well, this being a prime example of it.

What Gemini Said:

The history of the Cleveland Browns is a unique saga in American sports, defined by early absolute dominance, a heartbreaking mid-1990s relocation, and a modern era of rebuilding.

Founding and Early Dominance (1946–1962)

The franchise was founded in 1946 as a charter member of the All-America Football Conference (AAFC). The team was named after its first head coach, Paul Brown, a revolutionary figure who introduced many modern football staples, such as film study, the practice squad, and the draw play.

• AAFC Dynasty: The Browns won the league championship in all four years of the AAFC’s existence (1946–1949).
• NFL Transition: When the team joined the NFL in 1950, they immediately silenced skeptics by winning the NFL Championship in their first season. Led by quarterback Otto Graham, they reached the title game every year from 1950 to 1955, winning three of them.
• Integration: In 1946, Paul Brown helped break the professional football color barrier by signing Bill Willis and Marion Motley, a year before Jackie Robinson’s MLB debut.

The Jim Brown Era and Last Title (1957–1995)

After Otto Graham retired, the Browns drafted fullback Jim Brown in 1957, who is widely considered one of the greatest players in NFL history.

• 1964 Championship: Under coach Blanton Collier, the Browns defeated the Baltimore Colts 27-0 to win the 1964 NFL Championship. This remains the franchise’s last league title.
• The “Kardiac Kids” & Bernie Kosar: The team saw a resurgence in the late 1970s and 80s. Quarterback Brian Sipe led the “Kardiac Kids” to thrilling last-minute wins, and later, Bernie Kosar led the team to three AFC Championship games in the late 80s—all of which ended in painful losses (including “The Drive” and “The Fumble”).

Relocation and the “Deactivation” (1995–1999)

In a move that shocked the sports world, owner Art Modell announced in 1995 that he was moving the team to Baltimore.

• The Settlement: Following intense legal battles and fan protests, a unique agreement was reached: Modell would move his personnel to Baltimore to form a “new” team (the Baltimore Ravens), but the Browns’ name, colors, and 50-year history would remain in Cleveland.
• Hiatus: The franchise was officially “deactivated” for three seasons while a new stadium was built on the site of the old Cleveland Municipal Stadium.

The Expansion Era (1999–Present)

The Browns returned to the field in 1999 as an expansion team under owner Al Lerner.

• Modern Struggles: Since their return, the team has struggled to find consistent success, famously enduring a 0–16 season in 2017.
• Recent Years: The franchise has shown signs of a turnaround, reaching the playoffs in 2020 and 2023. The team is currently owned by Jimmy and Dee Haslam.

Summary of Championships

League	Years Won
AAFC Championships	1946, 1947, 1948, 1949
NFL Championships	1950, 1954, 1955, 1964

Citation: “Cleveland Browns History.” Gemini, Google, 24 May 2024,gemini.google.com/app.

AI feels like something that is just helping us, but the more I learn, the more I realized it can actually cause real problems if we were not paying attention. The concern I focused on is bias in AI, like how it can treat people differently based on race or the gender.

The most surprising thing I learned was how facial recognition systems are a lot less accurate for darker-skinned women compared to lighter skinned men. That is a mistake that can affect real life situations like criminal activity, job hiring, or security decisions. It made me realize that AI isn’t neutral but its trained on data created by humans. and if that data has any bias, the AI will reflect it.

Another thing is how important prompting is. If you ask a vague question, you get back a basic answer, but if you are specific and ask for multiple perspectives and answers, the responses are a lot better. I wouldn’t fully rely on AI for anything really important without checking it first because it can still sound confident even though it is wrong.

Overall, AI is useful and powerful, but it is not perfect. If we don’t question it, we could end up trusting systems that reinforce the same problems that we are trying to fix.