back

by dosinga·3y ago·view on hn ↗
Thanks, this is excellent feedback:

* The REPL is supposed to do autocomplete where it makes sense. If you enter A1. it should show you a list of things you can do there. That always feels better than asking for it explicitly. Let me know if that doesn't work.

* Yeah, code changes don't rerun the affected cells. It is something we've debated, and maybe we should? It feels like it could also cause unexpected things to happen. Hmm. Maybe we should though.

* This is tricky. Spilling from a function, doesn't fill the cells and then append does. Open for suggestions on how to make this better.

* We call the OpenAI API - they don't always answer these days :-(

* That's a good point. We should fix that!

3 comments
Excel has an option for formula calculation, automatic vs manual. The default is automatic. There is also a 'calculate now' button to use for if you have manual calculations set. I would expect that if automatic calculations set, then my code changes would trigger calculations on affected cells. I can't speak for Sheets, but most Excel users would probably expect that behavior.
Maybe? The closest analog in Excel would be to define a VBA function call it from a sheet, go back to the VBA editor, change the definition and see the function be run automatically everywhere with the new code. That does not happen, I just tried
For the tab completion, I think I tried typing the beginning a function name (hello) and then hit tab and it didn't complete.

> Yeah, code changes don't rerun the affected cells. It is something we've debated, and maybe we should? It feels like it could also cause unexpected things to happen. Hmm. Maybe we should though.

You know your customers better than I do. :). But generally when I use spreadsheets I expect bidirectional connections and updates.

> This is tricky. Spilling from a function, doesn't fill the cells and then append does. Open for suggestions on how to make this better.

I honestly don't know. Depends on your user's use cases. Even just for me I can see cases where I would want the cell updated and others where I wouldn't. Heck maybe even just ask the user "Do you want to update the cell?"

> We call the OpenAI API - they don't always answer these days :-(

That's fine, just tell the user it failed. :)

> We call the OpenAI API - they don't always answer these days :-(

Yikes. That's worrying. Is this rate limiting or API timeouts that need some retry logic wrapped around?

I think it might have been a quote issue actually ... But we should handle failures better, I agree
My worry was more about random failures on Open AI side (I'm building something with it). Thanks for sharing. I'd love to see a blog post on how you are handling the quoting issues.

Are you planning on sticking with Open AI or is it a stop gap until you train your own models?

>quoting issues Did you mean quota issues?

As for our own models, it depends on the use case. If you click the code assist and type "Insert a pie chart of the sources of income into B3", the Open AI models do a hard to beat (at this point) job. When it comes to fine tuning on a massive stack of existing spreadsheets, published models are a better fit.

Ah quota issues make more sense.

Thanks.