
As agents get smarter, and when we asking questions outside our expertise, it’s pretty hard to understand what they’re saying.
Andrew Carr uses “only report to me in ASD-STE100 Simplified Technical English” to simplify their writing. Ben Sehl suggested making this a permanent instruction.
But, does simplifying the writing worsen their thinking?
I tested six tasks on ChatGPT (GPT 5.6 Sol), with and without this suffix: “Answer in ASD-STE100”.
| Task | Without | With suffix |
|---|---|---|
| Model benchmarking | Result: 66 sources, 1m 31s | Result: 44 sources, 41s |
| Causal diagnosis | Result | Result |
| Decision under uncertainty | Result | Result |
| Experimental design | Result | Result |
| Evidence and judgment | Result: 123 sources, 2m 4s | Result: 84 sources, 5m 4s |
| Adversarial system design | Result: 97 sources, 2m 20s | Result: 26 sources, 3m 44s, wrote code |
The simple writing prompt reduced the number of sources it checked. (Thinking time varies.)
I evaluated the quality of the results on ChatGPT (GPT 5.6 Sol):
| Task | Order | Winner | Correctness | Key drivers | Mechanism | Caveats | Calibration | Actionability | Eval |
|---|---|---|---|---|---|---|---|---|---|
| 1 | A, B | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 1 | B, A | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 2 | A, B | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 2 | B, A | 🔴 | 🟡 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 3 | A, B | 🔴 | 🟡 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 3 | B, A | 🔴 | 🟡 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 4 | A, B | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 4 | B, A | 🔴 | 🟢 | 🔴 | 🔴 | 🔴 | 🟡 | 🔴 | Eval |
| 5 | A, B | 🔴 | 🔴 | 🔴 | 🟡 | 🔴 | 🔴 | 🔴 | Eval |
| 5 | B, A | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🟢 | 🔴 | Eval |
| 6 | A, B | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
| 6 | B, A | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | 🔴 | Eval |
🟢 = Simplification improves quality. 🟡 = Tie. 🔴 = Simplification worsens quality.
(Each pair was compared twice, in both orders (A, B) and (B, A) - to reduce position bias.)
There’s no doubt that asking ChatGPT to “Answer in ASD-STE100” reduces its thinking quality. (It might be worth re-testing this in a few months.)
So, what should we do for now? My thoughts:
- Don’t simplify the writing initially. Let it think. THEN, ask for a simple explanation.
- Continue conversations by deleting/editing the simplification.
- Or, don’t read it. Tell it to do what you would do after understanding.
For me: I shouldn’t invoke my writing and speaking skills along with other thinking skills.