In the few days since we wrote this post, Code Llama came out, including a flavor fine-tuned on 100B tokens of Python code - could be great for Grist, I look forward to re-running the benchmarks. This is an unexpected side benefit of using a common, popular language for Grist's formulas...
I appreciate the command first syntax, but maybe something like
; SELECT FROM sources ... WHERE / ORDER / ETC ... DATA fields ...
What if DATA was a reserved keyword for column identifiers and other data fields that would normally be immediately after SELECT so they could appear anywhere in the query syntax?
TLDR: Spreadsheets are the original low-code platform, but that little bit of code is still daunting. But for a spreadsheet with (1) formulas in a language with lots of training data (Python) and (2) an explicit relational structure that can be fed into a prompt, current large language models already look up to the job of taking care of the coding.
> are the original low-code platform, but that little bit of code is still daunting.
It may seem that this formula generation is just pushing Grist more into the no-code territory, that is, only for people who [add you favourite no-code crowd description here]
That is just not true.
- First, like all code assistants, you are still in full control. you can 100% edit the generated python formula to your likes.
- Second, and more generally, grist can be fully integrated with other solutions using its API (≈ backend) and its grids can be fully rewritten with custom html widgets (≈ frontend)
High-Code (and only if you need it), In, Back, and Front.
