AI cell generation #137
AI cell generation #137
Conversation
| autoFocus | ||
| onChange={(e) => setPrompt(e.target.value)} | ||
| placeholder="Write a prompt..." | ||
| className="flex min-h-[60px] w-full rounded-sm px-3 py-2 text-sm placeholder:text-muted-foreground focus-visible:outline-none pl-0" |
There was a problem hiding this comment.
Why not codemirror here so we can get auto-resize? Otherwise, we should make this a bit bigger and add a resize-none
There was a problem hiding this comment.
Why not codemirror here so we can get auto-resize?
Codemirror felt heavyweight for what is essentially a simple multi-line input
we should make this a bit bigger and add a resize-none
I like the resize, it's not too visible, but gives the user control should they be pasting a bigger prompt (unlikely, but possible). Why don't you like it?
| @@ -128,6 +128,25 @@ router.post('/generate', cors(), async (req, res) => { | |||
| } | |||
| }); | |||
|
|
|||
| // Generate a cell using AI from a query string | |||
| router.options('/sessions/:id/generate_cell', cors()); | |||
There was a problem hiding this comment.
Why not a websocket event?
There was a problem hiding this comment.
I don't have a super strong reason: I favor http when i need request/response semantics with a pending state. It's also less code.
Do you think this would be better as websocket?
|
|
||
| You are tasked with generating a Srcbook code cell for the user. | ||
|
|
||
| A Srcbook is a JavaScript notebook following a markdown-compatible format called .srcmd. It's an interactive and rich way of programming that follows the literate programming idea. |
There was a problem hiding this comment.
It's an interactive and rich way of programming that follows the literate programming idea.
These prompts feel too wordy IMO. This ^, for example, doesn't seem necessary.
There was a problem hiding this comment.
Yes, the prompts can definitely be iterated on. This made more sense in the generate srcbook one...I'll remove that particular sentence
Without evals, it's a bit of a nacked process. I am happy to remove this, but the next step for prompt iteration is to set up some basic evals. For this particular flow, the evals could programmatically test that we get valid srcbook code back based on a set of known user request examples.
Once we have that, we can start iterating with a bit more confidence.
Do you have data that suggests that shorter prompts work better? I am happy to iterate on the prompts and shorten them, but I think that the example is one of the most valuable parts. LLMs do very well when you give them examples (few shot). Are your concerns on cost? latency? |
Introduce the ability to create cells (code, markdown) with AI.
Below is a video of the MVP experience. A few things to note:
https://www.loom.com/share/91e4905a691f424db2791539ddfaf3e8?sid=683873e2-0fae-4f98-8019-063002ff3b47