I wonder how chroma collections compares to using Postgres partitioning. I haven't done this personally, but you should theoretically be able to add a `PARTITION BY collection_name` to have the same effect as sharding between chroma collections.
what place do vector-native databases have in 2025? I feel using pgvector or redisearch works well and most setups will probably be using postgres or redis anyway.
What other local databases did you consider, and why did you choose DuckDB?
Google tells me a lot of other people stream CDC pipelines to DuckDB, but I'm not familiar enough with it to know what makes it such a compelling choice.
My university's dorms suffered from no AC, black mold, cockroaches, and yearly floods. I couldn't imagine if teachers and TAs had to live in such slums in addition to being given such inadequate pay.
Providing housing is just treating a symptom - universities should fix this by paying their staff enough so they can even afford rent.
I think it's a pretty odd choice to make capital letters and ascenders tall in the name of legibility, especially for programming fonts. Much like bolding text, making capital letters relatively larger will make them stick out more in the code. But, do we need capital letters to be more noticeable?
* for ALL_CAPS constants, there's really no reason why constants should stick out above the rest of the code. In fact, as an invariant, it's probably less important when reading code.
* If you mix up PascalCase and camelCase, your linter or compiler will most likely catch it
Maybe capital letters should blend in more so syntax highlighting and "code shape" (eg indentation) will dominate --- which are much more important for understanding code at a glance than how we chose to arbitrarily capitalize some names.
I have already tried what you're doing, and it didn't perform well enough for me. I've been developing this project for a two years now. Its memory isn't going to fit in a single prompt.
I imagine that your AI chatbots aren't as cheap or performant as they can be with your potentially enormous prompts. Technical details aside, just like when talking to real people, it feels nice when they recall minor details you mentioned a long time ago.
I develop AI girlfriends.
I've struggled a lot with achieving natural-feeling recall. I've gone through a few iterations, but I'm currently experimenting with a knowledge graph/vector hybrid that uses an LLM to extract facts to build the graph. Both the performance and $ cost really hurt, but it truly does breathe life into AI characters.
I've seen a lot of the commercial products using the latest and most expensive models in the cloud to extract facts. Instead, I fine-tuned a local model on a gpt4-generated dataset, and it works surprisingly well. It will miss some connections but in practice I don't think it will be too noticeable.
The law is about "digital forgeries" in general - trying to pass off any computer-generated fake as authentic - but they're marketing this as just the "deepfake porn law" for voter popularity.
I think nothing less than a watermark or caption will suffice, but this is probably a problem that will have to be answered in courts.
I can't believe they would pass a law with such ambiguity. What do these people do all day?
If you read the actual law, it looks like this law pertains to "digital forgery" - computer-generated material that "falsely appears to be authentic". Does this mean that AI-generated porn is OK as long as it is explicitly labelled?
South Carolina is currently in a funny state where it's illegal to distribute non-consensual AI-generated porn of someone, but there aren't any laws against non-consensual revenge porn. Good news is that revenge porn laws have been adopted almost country-wide.