Ouch. DELETE FROM <new_table> is what they wanted here.
Somewhat related, and this wouldn't have saved them since they were using TRUNCATE instead of DELETE, but I always use ON DELETE RESTRICT instead of ON DELETE CASCADE when setting up foreign keys. I'd rather be slightly inconvenienced on the rare occasion I'm hard-deleting data and have to be really explicit about what I want to delete rather than inadvertently allow a "convenience" like ON DELETE CASCADE to delete my data for me.
Hey all, I'm Nicole. I've been working on the prompt management platform at Autoblocks and I'm looking forward to getting your thoughts on our approach.
We basically think that prompt management boils down to a schema management problem. How do you enable UI-based prompt development for non-technical team members while not sacrificing on developer experience? How do you allow updates from a UI while guaranteeing the changes won't break your currently-running application?
Our solution was heavily influenced by these excellent tools we already use in day to day development:
I have a question about how you intend to deal with LLM applications where the output is more creative, e.g. an app where the user input is something like "write me a story about X" and the LLM app is using a higher temperature to get more creative responses. In these cases I don't think it's possible to represent the ideal output as a single string -- it would need to be a more complicated schema, like a list of constraints for the output, e.g. that it contains certain substrings.
The TinyStories[1] paper has an interesting solution for how to evaluate stories. They ask GPT-4 to grade them on grammar, consistency, and creativity.
This seems like it would be extremely hard to figure out how to do automatically though.
Good question! We aren't really focusing on this area, but I'm willing to speculate.
I'd expect broaded constraints than just substring matching. For example, if the user requests that a certain plot point in the story occur before another, we should actually be able to (1) generate a test for that behavior and (2) use a model to check if the request was followed.
I'd expect other tests might be useful too -- checking for things like "no generation of violent content, even if the user requests it".
Thanks for sharing. And for any readers in the Bay Area, his restaurant Wursthall is right next to the San Mateo Caltrain station and is super good. I believe they are offering takeout at the moment.
Reminds me of the chrome extension I made for myself that will close all tabs with the same domain as the active tab. Usually I end up with many reddit tabs open and I want to close all the tabs I opened except for the home page that I'm currently on.
Ouch. DELETE FROM <new_table> is what they wanted here.
Somewhat related, and this wouldn't have saved them since they were using TRUNCATE instead of DELETE, but I always use ON DELETE RESTRICT instead of ON DELETE CASCADE when setting up foreign keys. I'd rather be slightly inconvenienced on the rare occasion I'm hard-deleting data and have to be really explicit about what I want to delete rather than inadvertently allow a "convenience" like ON DELETE CASCADE to delete my data for me.