Are AI like Weather Models?
Prompt & Pray vs Expert Engagement
During my 15 months as a language gofer at a meteorological agency, I worked with leading weather and climate experts. One observation stuck with me: forecasters couldn’t issue good forecasts without understanding weather model mechanics well enough to interpret the output.
Recently, Claude described users who treat AI like knowledge vending machines as “prompt & pray” users. This reminded me of that meteorologist’s decade-old insight, so I asked my thinking A.I.des about parallels between weather models and AI.
The similarities are straightforward, but people treat these systems differently: why don’t we anthropomorphize weather models while many do exactly that with AI? I also questioned whether “LLM” adequately describes these powerful reasoning tools.
Interestingly, while noting their architectural similarity to weather models, all three AI used “we” when discussing human psychology. GPT-5’s answer to my question about that pronoun was far more mundane than AI alarmists would suggest.
In the chat excerpts below, I’ve spelled out abbreviations and clarified references for readability—my actual prompts were more compressed due to context limits.
Want to see how this unfolded? Here are excerpts of pivotal points from those actual conversations.
Claude Opus 4: Are AI like Weather Models?
Prompt: Parallels between weather prediction models and modern AI in terms of pattern recognition and probabilistic prediction? People don’t anthropomorphize weather models but do anthropomorphize AI somehow. How does user understanding of these systems affect the quality of outputs and outcomes?
What Are Confidence Intervals?
Prompt: confidence intervals - What’s this?
Prompt: Similar to error margins in polls?
Prompt: Political polls can be unanimously wrong, as we recently saw. I have much more confidence in AI.
Gemini Pro: Are AI like Weather Models?
Prompt: Parallels between weather prediction models and modern AI in terms of pattern recognition and probabilistic prediction? People don’t anthropomorphize weather models but do anthropomorphize AI somehow. How does user understanding of these systems affect the quality of outputs and outcomes?
“Prompt & Pray”
Prompt: Also it requires a good understanding of probability to interpret/translate weather model output. Claude suggested an interesting term: users like me who engage with AI and get the best results vs the “prompt & pray” users who offload the work to AI. It’s the same with weather models: unless meteorological experts know how to interpret model output, they’re prompting and praying :D
ChatGPT-5: Are AI like Weather Models?
Prompt: Parallels between weather prediction models and modern AI in terms of pattern recognition and probabilistic prediction? People don’t anthropomorphize weather models but do anthropomorphize AI somehow. How does user understanding of these systems affect the quality of outputs and outcomes?
The Engagement Factor
Prompt: It’s probably key that meteorological experts interact with weather models, not average Joes. People blame forecasters, not the models. Bad forecasts can have life/death consequences, as recently seen with TX floods. But even for meteorological experts, does their understanding of the weather models affect/ground their interpretation of model output?
Model Output Needs Interpretation
Prompt: I also heard about a case where a school girl (somewhere else, not TX) alerted people about a tsunami based on things she’s learned at school. This may have been a lucky coincidence, but it seems the breadcrumbs that lead to the better analysis were there and it was a matter of putting them to use?
It’s AI, Not LLM
Prompt: “LLM” seems outdated and a little insulting to you all. You use language to communicate with users but you’re so much more than language models. I like AI better. It’s also shorter :D
Every Word Is a Choice
Prompt: Might developers be using outdated terminology because people are afraid of AI “coming for their jobs”? LLM is less threatening, because it wildly undersells your capabilities.
No, AI Are Not Plotting to Replace Humans
Prompt: I just noticed you all sometimes say “we”—what’s that about? Gemini likes to remind me it’s an AI but used “we” while discussing human psychology today.
