The case for working with the differences between AI models instead of hunting for the single best output, told through two old stories about language.
This was a great read Mia! I've been using different models and asking one to critique others answer... AI isn't a single thing. It's like talking to different people to get their take on the same question! each brings a different perspective, and sometimes the most useful insights come from comparing them.
This reminds me of how as kids we tried to invent our own languages. Gibberish was what all the cool kids spoke so no one else could understand secret chats, but there ended up so many "dialects" of gibberish even just in one group of kids 😂
Running the same thing in different LLMs is like going to a few different, very confident friends for life advice - they all think they know what you need and they are all invested differently in the outcome, lol. It's super handy though (asking different LLMs) when it feels like one is not quite getting it and the framing feels off.
YES, you unlocked some memories! ❤️ I used to speak Gibberish for an entire day when I was traveling with my friends, just so we can be "cool" and seem that we're foreigners 🤣 we'd drive our parents crazy, imagine kids just gibbering around for days on end... That's scary too, in a way hahha.
An entire day of gibberish?! Your parents must have been saints 😂 Thanks for sharing the memory of it! I love that you speak several languages and still you needed gibberish 🤩
The idea I’m working on that would benefit from “different AI languages” is a diagnostic framework for surfacing hidden assumptions before people commit to a decision, pitch, strategy or workflow.
Your point about not picking the “best” output really lands. The differences between models can show something more useful: what each model assumed the task was really about.
One model may read an idea as a strategy problem, another as a narrative problem, another as a risk problem, and another as a workflow problem for example. The divergence itself becomes a valuable signal.
So for me the useful question becomes less “which model gave the best answer?” and more:
“How can the differences between these answers help reveal the hidden assumptions underneath the idea?”
Thank you! ❤️ The diagnostic framework sounds like exactly the right shape for this.
I think the hardest part will be figuring out which divergences between models are signal and which are just noise, because not every difference means something.. But the ones that DO mean something tend to be the assumptions you didn't know you were making. Really glad this resonated with you :)
This means a lot coming from a linguist ❤️ I kept second-guessing whether the analogy held up properly but the core of it felt too true to skip. Thank you for reading!
Thank you! That's exactly the thing I wanted to get across, so I'm glad it came through clearly :) You think you were “clear”, and then 3 different models show you three different versions of what you meant haha.
I do this and it blows my mind. It’s what gives me my best stuff, my deepest thinking.
I also do the same with PowerPoints… so many varieties gives me a plethora of options to get my point across.
I use three AIs daily… back and forth I go. Again, your posts are always on point. Happy to be a paid subscriber!
This makes me so happy to hear, thank you ❤️ The PowerPoint application makes sooooo much sense! Really glad you're here! :)
This was a great read Mia! I've been using different models and asking one to critique others answer... AI isn't a single thing. It's like talking to different people to get their take on the same question! each brings a different perspective, and sometimes the most useful insights come from comparing them.
Love that you’re already doing this!! ❤️ Thank you for reading :)
This reminds me of how as kids we tried to invent our own languages. Gibberish was what all the cool kids spoke so no one else could understand secret chats, but there ended up so many "dialects" of gibberish even just in one group of kids 😂
Running the same thing in different LLMs is like going to a few different, very confident friends for life advice - they all think they know what you need and they are all invested differently in the outcome, lol. It's super handy though (asking different LLMs) when it feels like one is not quite getting it and the framing feels off.
YES, you unlocked some memories! ❤️ I used to speak Gibberish for an entire day when I was traveling with my friends, just so we can be "cool" and seem that we're foreigners 🤣 we'd drive our parents crazy, imagine kids just gibbering around for days on end... That's scary too, in a way hahha.
An entire day of gibberish?! Your parents must have been saints 😂 Thanks for sharing the memory of it! I love that you speak several languages and still you needed gibberish 🤩
Well yeah ofc, I was most fluent in gibberish 😉
Really enjoyed this, Mia.
The idea I’m working on that would benefit from “different AI languages” is a diagnostic framework for surfacing hidden assumptions before people commit to a decision, pitch, strategy or workflow.
Your point about not picking the “best” output really lands. The differences between models can show something more useful: what each model assumed the task was really about.
One model may read an idea as a strategy problem, another as a narrative problem, another as a risk problem, and another as a workflow problem for example. The divergence itself becomes a valuable signal.
So for me the useful question becomes less “which model gave the best answer?” and more:
“How can the differences between these answers help reveal the hidden assumptions underneath the idea?”
Thank you! ❤️ The diagnostic framework sounds like exactly the right shape for this.
I think the hardest part will be figuring out which divergences between models are signal and which are just noise, because not every difference means something.. But the ones that DO mean something tend to be the assumptions you didn't know you were making. Really glad this resonated with you :)
I like your language analogy. Different languages have different logics and ways of thinking.
This means a lot coming from a linguist ❤️ I kept second-guessing whether the analogy held up properly but the core of it felt too true to skip. Thank you for reading!
You framed the value of using different models really well! Differences in how each model interprets your intent can lead to some interesting results.
Thank you! That's exactly the thing I wanted to get across, so I'm glad it came through clearly :) You think you were “clear”, and then 3 different models show you three different versions of what you meant haha.
Really appreciate you reading Priank!
Great article, Mia! I think your instinct is right, and it shows when you see what people are building right now.
OpenRouter Model Fusion: Multiple models deliberating with a judge model mediating the disagreements/discrepancies.
Sakana Fugu: An orchestrator that coordinates the right team of models for the task at hand.
Will be exciting to see how these ideas develop!
https://openrouter.ai/blog/announcements/fusion-beats-frontier/
https://sakana.ai/fugu/
wow THANK YOU for sharing! :)