4 ms·

Thanks for the feedback! We aimed to make database migration easier and found that AI could really enhance this process by handling the complexities of differen

Thanks for the feedback! We aimed to make database migration easier and found that AI could really enhance this process by handling the complexities of different DB dialects. We'll also explore more deterministic approaches, as you suggested, using a common internal representation. The "Examples" link is being fixed—thanks for catching that!

teleforce 2y ago
Not a database expert but CUE can be used for generic schema design and validation, maybe it can also be used for deterministic database schema extraction and migration [1]. Unlike AI's non-deterministic techniques for example LLM approaches to NLP, CUE utilizes a well established alternative NLP deterministic approach namely Feature Structure [2], [3]. [1] Cue – A language for defining, generating, and validating data: https://news.ycombinator.com/item?id=20847943 https://news.ycombinator.com/item?id=20847943 [2] Feature structure: https://en.m.wikipedia.org/wiki/Feature_structure https://en.m.wikipedia.org/wiki/Feature_structure [3] The Logic of CUE: https://cuelang.org/docs/concept/the-logic-of-cue/ https://cuelang.org/docs/concept/the-logic-of-cue/
RadiozRadioz 2y ago
The different dialects have complexities, but they are they are known and explicitly defined. This task is natural to solve deterministically, and I'd say it's also important to do so - accuracy and consistency are paramount when migrating. I am having an extremely hard time understanding what benefits AI has here over a simple transpilation approach. Two benefits coming to mind are that AI is a good buzzword, and "shove it into ChatGPT, that'll do" takes less effort than building a proper transpiler.
- bal00ns 2y ago
  Like you said I imagine it was less effort. Less effort can be very valuable if you’re still producing something useful. Perhaps you could use a simple transpilation approach yourself if you think that’s a good idea.

teleforce 2y ago

Not a database expert but CUE can be used for generic schema design and validation, maybe it can also be used for deterministic database schema extraction and migration [1]. Unlike AI's non-deterministic techniques for example LLM approaches to NLP, CUE utilizes a well established alternative NLP deterministic approach namely Feature Structure [2], [3]. [1] Cue – A language for defining, generating, and validating data: https://news.ycombinator.com/item?id=20847943 https://news.ycombinator.com/item?id=20847943 [2] Feature structure: https://en.m.wikipedia.org/wiki/Feature_structure https://en.m.wikipedia.org/wiki/Feature_structure [3] The Logic of CUE: https://cuelang.org/docs/concept/the-logic-of-cue/ https://cuelang.org/docs/concept/the-logic-of-cue/

RadiozRadioz 2y ago

The different dialects have complexities, but they are they are known and explicitly defined. This task is natural to solve deterministically, and I'd say it's also important to do so - accuracy and consistency are paramount when migrating. I am having an extremely hard time understanding what benefits AI has here over a simple transpilation approach. Two benefits coming to mind are that AI is a good buzzword, and "shove it into ChatGPT, that'll do" takes less effort than building a proper transpiler.

bal00ns 2y ago
Like you said I imagine it was less effort. Less effort can be very valuable if you’re still producing something useful. Perhaps you could use a simple transpilation approach yourself if you think that’s a good idea.