> I had no issues pointing out Biscuit immediately but LLMs struggled even after extended analysis to design something similar even when it was aware of Macaroons.
I’m surprised that Google hasn’t integrated Gemini more natively into their workspace apps like Google Sheets or Google Slides, and has instead been developing a more ringfenced NotebookLM. I think they have some very basic integration in sheets but it feels like they’re keeping it intentionally subdued for now.
I suppose they want to take their time to get it right, but there are some killer use cases for LLMs in spreadsheets. Of course, the typical corporate idiot will make a giant mess of this when they use it in the wrong scenario… and I’m dreading the day I need to look at LLM-generated slides. But it still feels like Google is keeping Gemini in its own little box, rather than letting it seep more naturally into the rest of their products.
It's been increasing. They have the =AI function, generative tables, and a thing to read and "analyze" your stuff for you (though it generally says things you could probably guess). But I'm sure it'll get even better.
> Why do I even need to use the Unity3d editor for this? Why not store everything as code instead?
Some answers are... you need to recompile ever time you want to iterate on the data. You need to do a full reinstall if you were doing anything past play in editor. You need to do a week long store update if you actually shipped a game on an AOT only platform.
But yes, Unity assets don't support polymorphism like this. The data container needs to get massaged a bit.
In this case I'm still converting to assets so I can ship asset files in future updates, but yes play testing, while doable on the editor, has to be recorded in code as well, which may not be as accessible to a non-technical designer in the future, but that's for future me to solve (maybe at that time i'll finally have time to write a fully custom inspector for this) :D
I agree, this sounds very interesting and I'd like to apply to my projects. But without seeing the input and output it's up to me to creative think of abstract examples
I basically gave it the prompt i mentioned in the blog post and it then generated a final prompt. the final prompt is fitted against the data i have but if you're interested this is the final prompt
It's been a long minute since my Unity days, but I'm pretty sure what OP wanted was ScriptableObjects with a custom Inspector
If you setup asset serialization to use text, the ScriptableObjects map to semi-editable YAML files (semi- editable because there are some internal members that show up, but nothing an LLM couldn't handle)
I already use ScriptableObjects with a custom inspector that outputs to YAML files. the problem was that with increasing use of nested generics+components I would need to spend alot of time customising the inspector. I rather not spend time on tooling and focus on writing game logic.
The downside of that is that cards often have to hook into custom functions that interface with the engine (unless they're sufficiently boring). You can still go with ScriptableObjects if you're willing to e.g. write those functions in Lua and have an interpreter layer, but for all but the largest projects, making each card its own class works well.
Scriptable Objects are exactly what I would use in this case. There's even a whole host of fancy tools to work with them and import/export to csv or other file formats.
LLMs are not intelligent because they're pattern matching and those patterns are always at the tip of the context with no recall to any other point in time.
Also, humans are not intelligent because they forget.
That's it! The article beautifully summarizes how I interact with LLMs. Listing state of the art solutions and reformatting data is an excellent use case.
> I had no issues pointing out Biscuit immediately but LLMs struggled even after extended analysis to design something similar even when it was aware of Macaroons.
I’m surprised that Google hasn’t integrated Gemini more natively into their workspace apps like Google Sheets or Google Slides, and has instead been developing a more ringfenced NotebookLM. I think they have some very basic integration in sheets but it feels like they’re keeping it intentionally subdued for now.
I suppose they want to take their time to get it right, but there are some killer use cases for LLMs in spreadsheets. Of course, the typical corporate idiot will make a giant mess of this when they use it in the wrong scenario… and I’m dreading the day I need to look at LLM-generated slides. But it still feels like Google is keeping Gemini in its own little box, rather than letting it seep more naturally into the rest of their products.
I'm happy they're cautious with spreadsheets. Wrong numbers can really mess up people's lives.
That said, I saw an add for Gemini for Google Sheets today.
It's been increasing. They have the =AI function, generative tables, and a thing to read and "analyze" your stuff for you (though it generally says things you could probably guess). But I'm sure it'll get even better.
They saw copilot be… copilot.
good
> Why do I even need to use the Unity3d editor for this? Why not store everything as code instead?
Some answers are... you need to recompile ever time you want to iterate on the data. You need to do a full reinstall if you were doing anything past play in editor. You need to do a week long store update if you actually shipped a game on an AOT only platform.
But yes, Unity assets don't support polymorphism like this. The data container needs to get massaged a bit.
In this case I'm still converting to assets so I can ship asset files in future updates, but yes play testing, while doable on the editor, has to be recorded in code as well, which may not be as accessible to a non-technical designer in the future, but that's for future me to solve (maybe at that time i'll finally have time to write a fully custom inspector for this) :D
Use spreadsheet to code reminded me of SPL https://github.com/SPLWare/esProc/wiki/A-programming-languag...
Would be nice to see the exact prompt, model used, and results - it's not exactly top secret data.
For reference this is the final prompt, though the initial meta prompt was included in the post
https://danieltan.paste.lol/example-battler-data-final-promp...
I am using claude sonnet 4
I agree, this sounds very interesting and I'd like to apply to my projects. But without seeing the input and output it's up to me to creative think of abstract examples
I basically gave it the prompt i mentioned in the blog post and it then generated a final prompt. the final prompt is fitted against the data i have but if you're interested this is the final prompt
https://danieltan.paste.lol/example-battler-data-final-promp...
It's been a long minute since my Unity days, but I'm pretty sure what OP wanted was ScriptableObjects with a custom Inspector
If you setup asset serialization to use text, the ScriptableObjects map to semi-editable YAML files (semi- editable because there are some internal members that show up, but nothing an LLM couldn't handle)
I already use ScriptableObjects with a custom inspector that outputs to YAML files. the problem was that with increasing use of nested generics+components I would need to spend alot of time customising the inspector. I rather not spend time on tooling and focus on writing game logic.
The downside of that is that cards often have to hook into custom functions that interface with the engine (unless they're sufficiently boring). You can still go with ScriptableObjects if you're willing to e.g. write those functions in Lua and have an interpreter layer, but for all but the largest projects, making each card its own class works well.
Scriptable Objects are exactly what I would use in this case. There's even a whole host of fancy tools to work with them and import/export to csv or other file formats.
Humans are not intelligent because they forget.
LLMs are not intelligent because they're pattern matching and those patterns are always at the tip of the context with no recall to any other point in time.
Also, humans are not intelligent because they forget.
That's it! The article beautifully summarizes how I interact with LLMs. Listing state of the art solutions and reformatting data is an excellent use case.