In partnership with
Yesterday I used Anthropic’s new Fable 5 model to build a fantastic physics simulator for our senior Physics students. It was genuinely good and it would have taken me far longer to create using Opus or Codex (and would not have looked anywhere near as good). Today I tried to build a biology one for a school field trip. Fable 5 blocked it, making me use Opus 4.8 instead.
I teach biology, chemistry, physics and psychology to high school students, and I'm Director of Learning Innovation at two schools here in New Zealand. Part of the role is showing staff and students how to actually use AI, alongside the science of how people learn. I pay for Claude Max 20x, Codex and Gemini myself (no school reimbursement) and it is certainly not cheap. So I was glad to be able to test Fable 5 over then next 14 days.
Fable 5 has been the standout of all the models so far. The speed, and the quality of what it helps me build, have been well ahead of my Opus 4.8 and Codex usage.
For everything, that is, except biology.
Of course, I get the bioweapons concern. If a model holds a hard line on anything that edges toward engineering a pathogen, all well and good. I don't want it doing that either.
What I don't understand is why ordinary, school-level biology (as well as ecology) sits on the wrong side of it.
Here is the example that stopped me. I'm taking a class to Mistletoe Bay in the Marlborough Sounds for a Year 13 ecology trip. We sample the intertidal zone: crabs, chitons, molluscs, and where they sit on the shore in relation to the tide. I wanted an interactive simulator so students could explore how organisms are distributed across a shore, and how you'd investigate that with line transects, quadrats and the rest. Being prepared is a good thing and a simulator will definitely help, and I have not been able to create one programmatically until this year.
This was my prompt. Read it and tell me where the danger is (feel free to adapt and use yourself ).
Fable, I am planning resources for our upcoming Biology camp at Mistletoe Bay in the Marlborough Sounds. We have already created learning resources to support students on the journey there, and I now want to explore whether we could create an interactive ecological simulator that helps students understand how to sample organisms in the intertidal zone.
The simulator should model how different organisms are distributed across the shore and how students might investigate those patterns using ecological sampling techniques.
To make the simulator biologically realistic, please research the following:
- The tidal patterns at Mistletoe Bay, Marlborough Sounds, during the week of 27 June to 3 July 2026.
- The true crabs and false crabs that may live in this area.
- The chiton species that may live there.
- The shellfish and other molluscs that may live there.
For each species or group of organisms, I need information about:
- where they are likely to live on the shore
- how they move, if they move
- their preferred substrate
- their tolerance of exposure, wave action, salinity, temperature, and desiccation
- relevant biotic factors such as competition, predation, grazing, shelter, and food availability
- relevant abiotic factors such as tide height, substrate, moisture, light, temperature, and wave exposure
Using this information, devise a simulator that could model species distribution across the shore. The simulator should allow students to explore how patterns may change in relation to:
- tide height and tidal timing
- substrate type
- exposure time
- competition
- predation
- shelter availability
- moisture and desiccation risk
- other relevant ecological variables
The simulator should also show how students could investigate these patterns using a range of ecological sampling techniques. Include techniques that would be practical for school students at Mistletoe Bay, such as:
- line transects
- belt transects
- quadrat sampling
- stratified sampling across shore zones
- random sampling
- sampling under rocks, where this is ethical and safe
- comparing exposed and sheltered sites
- comparing different substrate types
For each sampling technique, explain:
- what the method involves
- what kind of ecological question it is suitable for
- what data students would collect
- what the strengths and limitations are
- whether it is practical for Year 13 Biology students in the field
- any safety, ethical, or environmental issues
The students will also need to think carefully about confounding variables. Please include guidance on how they could identify and control possible confounding variables. For example, if students propose that a distribution pattern might be related to magnetic field variation, explain whether this is biologically plausible, how they might test it, and whether comparing Mistletoe Bay with a similar site in a different magnetic location would be practical or necessary.
I also need a student-friendly explanation of relevant statistical tests. The students do not need a detailed mathematical explanation of how each test works, but they do need to understand:
- what kind of question each test answers
- what type of data they need to collect
- how they would organise the data
- when they might use a chi-squared test
- when they might use ANOVA
- when they might use correlation or regression
- what a p-value means
- how to interpret statistical significance carefully
- how to avoid overclaiming from their results
We are likely to use Jamovi for the statistical analysis, so please explain how the simulator and data collection could link to Jamovi. It may be useful to create a separate statistics explainer for students, including worked examples using ecological data from the simulator.
The final aim is to support students as they design, carry out, analyse, and write up their Biology investigation over the school holidays.
Please also research and recommend the best programmatic solution for building the simulator so that students can use it easily on their own devices. We have previously created apps using Vercel, but I am not sure whether this is the best option here. Consider whether a React-based web app, a Vercel deployment, or another approach would be most appropriate.
In your response, please include:
- A biological research summary for the Mistletoe Bay organisms.
- A proposed ecological model for the simulator.
- Suggested sampling techniques and how they would appear in the simulator.
- Guidance on practical fieldwork design.
- Confounding variables students should consider.
- Recommended statistical tests and how students would use Jamovi.
- A recommended technical approach for building the simulator.
- Any questions you need answered before development begins.
There's definitely no genetics, molecular biology or microbiology. Just crabs, seaweed, a tide table and some statistics.
It was blocked. Here is exactly what it came back with:
Read that again.
The message itself says it may be flagging safe, normal content, and that a school trip about crabs is part of the price for Mythos-level capability somewhere else. Then it switched me to Opus 4.8, and I carried on with the model I'd already found weaker for this kind of build.
With the physics simulator, the result was excellent. With my biology, I never got to see what it would have looked like, or whether it would even have been any good. That is the entire reason for testing Fable 5 with it. And I couldn't, because it was about a field trip involving crabs, chitons and molluscs.
That doesn't seem fair. I could live with it if biology were the only casualty. It isn't. It's the same story across chemistry and psychology. Building tools that help students understand concepts is a good thing.
Just not, apparently, in three of the subjects I happen to teach.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
This is where it stops being about my frustration. The bottleneck here isn't capability. The model can obviously do the work. The bottleneck is a safety filter that can't tell a bomb from a beach. The cost of that filter doesn't fall evenly. If your work touches biology, chemistry or psychology, you are quietly locked out of the best models, presumably while paying the same as everyone else for less.
Ethan Mollick made a very interesting point in February, and reshared it this week as being "more true than ever":
Basically, in any career people will be wanting access to the best AI models to be efficient in their role and will want access via a token budget. If that's where we're heading, then anyone whose work sits in biology, chemistry or psychology is starting a step behind, on tools that won't build for them what they will happily build for a colleague in another field. If you are in the blocked fields, you will be judged unworthy and will not have the golden tokens.
So how could Anthropic solve this?
One obvious route should be accreditation. Give teachers, and others who can show they are using biology or chemistry to help people learn, a way to verify that and have the filter eased accordingly. A registered teacher building an intertidal sampling simulator should not be a threat model. Treating them as if they are comes with a cost of its own, which I would argue is potentially more dangerous.
This is because the real version of this problem isn't me and one field trip. If the students in front of me get worse tools for biology, chemistry and psychology than they get for every other subject, that becomes a quiet reason to study something else. We would be steering young people away from exactly the subjects we keep saying the future needs, right up until the day a Project Glasswing company finally hands them the useful tools. This is harming the advancement of science by creating an AI underclass.
Judging by the comments on X in response to a request by the journal Nature, I’m not the only one in this position:
Finally, given what I have discussed today, the beautiful launch video for Fable 5 strikes me as being just a tad tone deaf - given the amount of biology on display!
Anyhow, for the moment my biology work will have to be a second-class citizen compared to my other work.
I’m interested in whether you have had similar issues and what you are doing about it.
Until next time, Ngā mihi
Eliot