@samgoodwin89 I guess we are in the age of personal software now ! I would never think of building an app like that without ai but now…
English
mathis
174 posts
mathis
@mathisob
Working at https://t.co/kbKJYf98zX for local newsrooms. Building https://t.co/dZ8BoXCni6 in public
Paris, France Присоединился Haziran 2009
354 Подписки42 Подписчики
@ryanflorence @SurelyAutomata Funny I came up with a similar solution but you don’t really need to do breadth first if you’re going to save the level of each node in a separate structure. You can do depth first without a queue and get the same result (with a bit less complexity)
English
@SurelyAutomata alright, back at my computer with the update
I keep the level with the node in the queue and added a function to get all the branches if you want to process an entire branch at once instead of just walking the tree
English
Not a CS grad (econ) and have never done leetcode stuff. I tried this just for fun. Turned off copilot.
I flailed for about 5 minutes trying to call a recursive function, realizing half way through each attempt it was going to do depth first
Then I remembered while loops
vik@vikhyatk
i have a very simple question i ask during phone screens: print each level of a tree on a separate line. 90% of CS grad candidates just can't do it. someone needs to investigate these universities.
English
the results of this are so surprising i think i'm just gonna conclude you're all liars
dax@thdxr
if your primary project for work is a nextjs app is it
English
@thomasankcorn You can defer if you use lambda function urls and streaming (or with extensions but DX is not great) aws.amazon.com/blogs/compute/…
English
@samgoodwin89 Yeah, I got to use python more for data tasks but it was much more pleasant that I expected thanks to copilot and chatgpt.
English
The one area that SST has been historically weak in is having a plugin ecosystem.
These were tricky partly cos of how we were using CDK and partly cos you needed to understand CDK/CFN.
We'll share details soon but with ❍ it'll be a lot easier to build and distribute a plugin.
English
@jayair Yes this makes sense when you really want to keep your data in a database that has no vector search built in like dynamo but many search engines / databases are adding vector search and it makes a separate vector database redundant. I ditched pinecone when typesense added vectors
English
Our little AI demo went viral over the weekend so I wanted to take a minute to share why we built it.
I also have a slightly different take on these LLMs.
*cough* AI is the future of UI *cough*
Let's do a thread!
dax@thdxr
we’re re-enacting the plot to silicon valley with LLMs we’ve been gifted the best way to search over compressed data ever invented but everyone is building the same chatbot over and over we built a 100% open source demo of something more practical check the thread
English
@jayair I think this setup could be improved if you could retrieve the actual data you are searching for instead of just retrieving the ids and then getting the data from dynamo with a batchGet. Vector search should just be one way to query your data, not a completely separated system.
English
@thdxr I think in most cases you want your vectors to be stored in the same database as the rest of your data. In this case storing movies in the same rds table would save you from doing a batchGet in dynamo after retrieving the vectors (that’s why I moved away from pinecone)
English
we've open-sourced this whole app so go explore it - i think a lot of products can benefit from this kind of thing
it's all very early but hopefully you can see the potential
PRs welcome
github.com/sst/demo-ai-ap…
English
