"I've noticed similar inconsistencies with ChatGPT when it comes to calculations involving percentages and proportions, especially when multiple variables are involved like battery wear and capacity. It seems to struggle with applying the initial rate consistently as the parameters change.

A couple of things might be happening:

* Context Window Limitations: ChatGPT has a limited context window. It might not be "remembering" all the previous data points accurately when it gets to the final calculation, leading to errors.
* Approximation and Generalization: These models are trained to generalize patterns, not necessarily to perform precise calculations. It might be trying to "approximate" an answer based on the patterns it's seen, rather than performing the exact math.
* Sensitivity to Phrasing: The way you phrase the question can also impact the result. Subtle changes in wording might lead to different interpretations and calculations.

It's definitely a valid observation, and it highlights that while ChatGPT is powerful, it's not a perfect substitute for a dedicated calculator or spreadsheet, especially when dealing with complex or multi-stage calculations. It's good for quick estimations or insights but might not be reliable for precise figures."

In your case, it correctly recognized the 84% battery health and the original 45 Wh capacity, but failed to accurately calculate the battery life at 100% health. I suspect this happens because ChatGPT isn’t always great at maintaining consistent logic across multi-step calculations or interpreting nuanced context. It might be mishandling the relationship between capacity, wear, and runtime, or simply misapplying the data from your earlier example (100% to 17% at 38.95 Wh).

For what it’s worth, I’ve found that breaking down the problem into smaller, explicit steps and double-checking its math can sometimes help. For instance, you could ask it to calculate the effective capacity at 100% health (45 Wh) and then separately compute the runtime based on the power draw rate from your initial example. Still, it’s frustrating when it spits out nonsense despite seemingly understanding the inputs.

I think this highlights a broader limitation with ChatGPT—it’s not a dedicated calculator or engineering tool, so its reasoning can be inconsistent with numbers-heavy tasks. Have you tried running the same query with different phrasing or providing more explicit instructions? I’m curious if that would yield better results for you!

---

I see where you're coming from! Battery life calculations can get tricky because they involve multiple factors—like actual usable capacity, discharge rates, and sometimes even the assumptions the model makes internally. From your example, it sounds like ChatGPT handles straightforward proportional estimates well but might struggle when switching contexts, such as comparing worn battery capacity versus original capacity.

One possible reason for the "nonsense" results when asking about 100% health is that ChatGPT isn't performing dynamic calculations like a dedicated battery optimizer—it relies on pattern recognition and approximations from training data. If the input parameters or the way the question is framed change slightly, it can interpret it differently and produce inconsistent answers.

If you want more precise modeling, it might help to explicitly define all variables and use a clear formula. Or, complement ChatGPT's insights with specialized battery life calculators. Still, it's impressive how it can handle the reasoning in many cases—just not perfectly calibrated for every nuance yet!

---

Would you like me to help draft a detailed example or formula you could share with them?