Anthropic recently released something that looks more polished but follows the chat paradigm. [1]
As a builder of something like that [2], I believe the future is a mix, where you have chat (because it's easy to go deep and refine) AND generate UIs that are still configurable manually. It's interesting to see that you also use plotly for rendering charts. I found it non-trivial to make these highly configurable via a UI (so far).
Thank you for open sourcing so we can all learn from it.
Since Data Formulator performs data transformation on your behalf to get the desired visualization, how can we verify those transformations are not contaminated by LLM hallucinations, and ultimately, the validity of the visualization?
After giving it a whirl I'm a little underwhelmed, but maybe I'm using it wrong. I'm getting less consistent results than if I prompted GPT4-o for a Vega graph after providing it with the documentation.
I rather like this idea. Apologies however for my cynicism in advance. I suspect it'll die due to human concerns. I've seen many reports recently which are just plain and utterly wrong written in dashboards by vendors and internally. The veracity of the results is mostly based on the human driving it and validating the methodology and the competent ones are apparently rather rare. This serves to give it to humans who are even worse at the job than the current ones.
Anthropic recently released something that looks more polished but follows the chat paradigm. [1]
As a builder of something like that [2], I believe the future is a mix, where you have chat (because it's easy to go deep and refine) AND generate UIs that are still configurable manually. It's interesting to see that you also use plotly for rendering charts. I found it non-trivial to make these highly configurable via a UI (so far).
Thank you for open sourcing so we can all learn from it.
[1] https://news.ycombinator.com/item?id=41885231 [2] https://getdot.ai
Here is the link to one of the prompts. It seems like all the LLM tasks are in the agents directory: https://github.com/microsoft/data-formulator/blob/main/py-sr...
Some of these "agents" are used for surprising things like sorting: https://github.com/microsoft/data-formulator/blob/main/py-sr... [this seems a bit lazy, but I guess it works :D]
Since Data Formulator performs data transformation on your behalf to get the desired visualization, how can we verify those transformations are not contaminated by LLM hallucinations, and ultimately, the validity of the visualization?
We can’t. Without the driver this car runs on probability. And that all. A capable operator is still needed in the loop.
Definitely looks like something that could save me, and others, allot of time. Thanks for sharing!
After giving it a whirl I'm a little underwhelmed, but maybe I'm using it wrong. I'm getting less consistent results than if I prompted GPT4-o for a Vega graph after providing it with the documentation.
way cool! I hope to take it for a spin tomorrow!
Q: Does your team see potential value in a DSL for succinctly describing visualizations to an LLM as Hex did with their DSL for Vega-lite specs [1]?
[1]: https://hex.tech/blog/making-ai-charts-go-brrrr/
Wow, that's pretty cool! I think there are potential -- current LLMs are not that good on VegaLite when I ask it to edit the script :)
Thanks for sharing and provide open source version! This is great!
I rather like this idea. Apologies however for my cynicism in advance. I suspect it'll die due to human concerns. I've seen many reports recently which are just plain and utterly wrong written in dashboards by vendors and internally. The veracity of the results is mostly based on the human driving it and validating the methodology and the competent ones are apparently rather rare. This serves to give it to humans who are even worse at the job than the current ones.