AI – Higher Education – Innovation

OpenAI has released ChatGPT 5, its most advanced model yet. The reaction – mixed..!

While some people are dismayed at the loss of previous versions (4o, o3), and lament the forfeiture of user choice, there’s no denying that version 5 is a step up in LLM capabilities, including coding.

The choice of model has now been taken out of the hands of the user, for reasons only known to OpenAI – although many suspect multiple models are becoming costly to run. ChatGPT now decides the best model that it deems fits the desired output.

I’m going to be honest – I have no clue how to code. I wouldn’t know where to start, least of all develop an application that I could use in my class to aid student learning. I have lots of ideas of what I might like to do but no idea how to create an end product without enlisting the help of an expert programmer.

I teach a first year undergraduate module covering Human Genetics and Evolution and struggle to convey concepts such as ecological dynamics and environmentally stable strategies. I’ve seen some nifty web-based apps that allow different organisms to interact and evolve over time but none really deliver what I’m looking for.

I turned to ChatGPT 5 for help. Could the model live up to the hype?

Collaboration with ChatGPT

My idea was to create a “Survival Island” where students could tinker with parameters, watch ecosystems react in real time, and through trial and error understand the balances that make a system evolutionarily stable. The aim was to create a vivid, browser-based simulation where plants, herbivores and predators interact according to defined biological rules.

Could, through a simple prompt, ChatGPT 5 deliver what I was after?

The answer – yes – but only after a few failed attempts and re-engineering of my prompts.

This was the original prompt:

The initial output was a draft outline of the concept and game loop scenario – text only. Therefore:

It certainly generated the code but I didn’t know what to do with it. I needed something frictionless and immediately downloadable. After a few rounds of back and forth we ended up here:

This is what it produced, which on first pass looked rather impressive and the functionality seemed to be very usable. ChatGPT named this version ‘Island of Adaptation’, which I thought apt. However, the functionality was poor and the simulation not constrained to sound biological principles:

I asked ChatGPT 5 to adjust the configuration of the visual window and render in 3D. This is what it produced:

Again, the simulation was poor, as organisms wandered off the landmass and into the sea..! There was no interaction between species either. Improvements were required.

Although ChatGPT is considered to be superior in long chains of tasks, it still lost the thread after multiple rounds of iteration. I’m left wondering if this is due to poor selection of model mid chain, which occurs behind the scenes remember, rendering the chain unresponsive to further commands.

Asked to improve the “Survival Island” simulation, ChatGPT 5 would sometimes update the code but render the application unusable or missing vital elements.

Far from perfect and therefore time for a slightly different strategy.

The final output – pretty good!

The best solution was to start a new chat and re-prompt ChatGPT 5. Unsurprisingly the output and format of the simulation was completely different. I did this two or three times until I stumbled on a form and simulation I considered met my expectations and needs. This was my starting prompt and a follow on prompt:

Through a further iterative process, I eventually turned that vision into a polished, interactive HTML5 simulation complete with live charts, parameter sliders, and now, thanks to the refinements, on-screen instructions and real-time Hardy–Weinberg (H-W) allele frequency calculations.

The final output was a comprehensive learning environment that enables my students to explore population ecology and evolutionary stability. The video below shows it capabilities (you can set the speed and adjust the parameters for each species, such as reproduction rate, speed of movement, energy requirements, etc.). Plus, there are options to randomly mutate the populations, force drought conditions or force hungry predators.

The first version of Survival Island already had the core features:

• Three trophic levels – plants (green), herbivores (blue dots), predators (red triangles) – each with adjustable biological parameters.
• A real-time frequency chart tracking populations.
• A flexible slider control panel to adjust growth rates, carrying capacities, metabolism, reproduction thresholds, and more.
• Pre-set scenarios like ‘Stable-ish’, ‘Drought’, and ‘Hungry Predators’.

However, I wanted to make it more student-friendly and scientifically rich.

The population frequency table updates in real time as does the Hardy-Weinberg equation for each species. It calculates p, q, p², 2pq and q² values for each species, plus expected genotype counts and species n-numbers. Students can see how allele frequencies might shift with population changes, adding a minor genetics dimension to the ecology.

You can pause and restart at any time and there’s a function to download a CSV file of the data.

There’s a drop down instruction tab for students to explore as well, describing each parameter for each species.

The ultimate goal is to achieve ecological equilibrium or an evolutionary stable state (ESS). Here’s the video:

The next job is to test this with students in class.

Why Simulate Evolution?

Evolutionary dynamics are about change over time in response to environmental pressures and interactions between species. But the ‘time’ in evolution can span generations.

Simulations compress those timescales, allowing students to see patterns that would otherwise take hundreds or thousands of years to emerge in nature.

In Survival Island, each run of the simulation is like a unique “mini-world,” with its own unfolding story and evolutionary history:

• Will the predators overhunt and starve?
• Can the plants regrow fast enough to sustain herbivores?
• What happens when a sudden change mimics a drought or a surge in predator speed?

Evolutionary processes happen over large timescales and in ways that are hard to witness directly. Survival Island brings them into the classroom.

How Students Can Use Survival Island?

The educational benefits come from active experimentation. Here’s how students might work through a lesson using the app:

1. Establish a Baseline – start with the ‘Stable-ish’ pre-set. Read through the instructions to understand the sliders. Let the simulation run and watch.
2. Explore a Single Parameter – change just one variable. For example, plant regrowth rate. Increase it slightly and observe whether herbivore numbers grow. Lower it and watch if herbivores decline. This isolates cause and effect.
3. Look for an ESS – challenge students to find settings where all three populations survive for 200 ‘days’ without extinction. Discuss why such states are stable and why they can be resistant to invasion by a mutant strategy.
4. Connect Ecology to Genetics – open the H-W panel and note that p=0.50 at the start. Let the sim run under different conditions and see if the p-value drifts, then ask what real-world factors could cause such changes?
5. Compare Scenarios – export CSV data from two contrasting runs. Plot them outside the app and compare. Which parameters gave longer stability? Which led to faster crashes?

By the end, students aren’t just observing outcomes, they’re forming hypotheses, running experiments and interpreting results.

Reflections on AI Collaboration

This project showcases what happens when educational requirements meet AI-assisted coding.

With the core simulation solid, I can now think about expansion and improvement to make Survival Island a richer simulation, based on student feedback and learning needs.

No coding was required, just carefully thought text prompting. Input your idea with a few considered principles or rules, and watch ChatGPT 5 do the rest.

Yes, there were long-chain dropouts and aborted attempts but not many. From prompt to output took less than an hour for each run.

What about ChatGPT 5?

I’m not a coding expert or programmer, therefore the ease in which text-to-application was achieved was certainly impressive. No coding expertise required, just a relatively simple prompts with a few iterations to modify the functionality of the platform.

I’m also aware that other LLMs are highly capable of coding, with tools such as Claude Opus 4 (Anthropic) that is often a preferable option for developers. However, as ChatGPT 5 is now the flagship OpenAI model (and free to use), this may become the choice for many who are not willing to pay for Pro or Enterprise level access to more sophisticated tools.

ChatGPT 5 offered a powerful resource to develop a relatively sophisticated teaching tool.

All you need is a little patience and an idea!